Micchi et al. (2020) - Improvements with data augmentation and alternative representation.

The code is running after I met with Gianluca and Mark.

Now it is time to make a contribution and improve the performance.

Doing experiments with the spelling_compressed representation compared to the spelling_bass representation.

Important note: I wasted a lot of time figuring out that my environment (python packages) changed and they dramatically changed the results. Possibly new music21 changes introduced with v3.7.0 broke the processing of scores.

For now, I will not investigate the root cause of this, the takeaway is: USE THE SAME GOD DAMN VIRTUAL ENVIRONMENT THROUGHOUT THE WHOLE EXPERIMENT, UNTIL PAPER SUBMISSION. FREEZE THAT ENVIRONMENT, MAKE IT PUBLICLY AVAILABLE.

Rant over.

Now into the specifics of what I am doing.

Here is the environment, by the way:

absl-py==0.11.0
astunparse==1.6.3
cachetools==4.2.1
certifi==2020.12.5
chardet==4.0.0
cycler==0.10.0
flatbuffers==1.12
gast==0.3.3
google-auth==1.24.0
google-auth-oauthlib==0.4.2
google-pasta==0.2.0
grpcio==1.32.0
h5py==2.10.0
idna==2.10
joblib==1.0.0
Keras-Preprocessing==1.1.2
kiwisolver==1.3.1
Markdown==3.3.3
matplotlib==3.3.4
more-itertools==8.6.0
music21==6.5.0
numpy==1.19.5
oauthlib==3.1.0
opt-einsum==3.3.0
pandas==1.2.1
Pillow==8.1.0
protobuf==3.14.0
pyasn1==0.4.8
pyasn1-modules==0.2.8
pyparsing==2.4.7
python-dateutil==2.8.1
pytz==2021.1
requests==2.25.1
requests-oauthlib==1.3.0
rsa==4.7
scipy==1.6.0
seaborn==0.11.1
six==1.15.0
tensorboard==2.4.1
tensorboard-plugin-wit==1.8.0
tensorflow==2.4.1
tensorflow-estimator==2.4.0
termcolor==1.1.0
typing-extensions==3.7.4.3
urllib3==1.26.3
webcolors==1.11.1
Werkzeug==1.0.1
wrapt==1.12.1
xlrd==1.2.0

Starting from the reported models

Running the spelling_bass_cut input with the gru model. Using gru simply because it has less parameters and it's faster to train. All experiments will run with gru and once I get an idea of the performance of my ideas vs Micchi et al., I will move to conv_gru
Restricting myself to the chen-su dataset. Not touching the test set. Training on train, validation on valid. Not touching test until the very end. If I can get the model to be better than Micchi et al., it will be better over the validation already.
First, no data augmentation. Data augmentation with my spelling_compressed is harder than it is with spelling_bass. I need to know how both do before data augmentation.

python train.py --model 2 --input 4

Epoch 81/100
 1/12 [=>............................] - ETA: 6s - loss: 2.8467 - key_loss: 0.2469 - degree_1_loss: 0.2610 - degree_2_loss: 0.5208 - quality_loss: 0.5484 - inversion_loss: 0.7494 - root_loss: 0.5203 - key_accura 2/12 [====>.........................] - ETA: 5s - loss: 2.9345 - key_loss: 0.2669 - degree_1_loss: 0.2807 - degree_2_loss: 0.5488 - quality_loss: 0.5653 - inversion_loss: 0.7518 - root_loss: 0.5210 - key_accura 3/12 [======>.......................] - ETA: 4s - loss: 3.0427 - key_loss: 0.2843 - degree_1_loss: 0.2905 - degree_2_loss: 0.5794 - quality_loss: 0.5938 - inversion_loss: 0.7586 - root_loss: 0.5361 - key_accura 4/12 [=========>....................] - ETA: 4s - loss: 3.1452 - key_loss: 0.2993 - degree_1_loss: 0.2978 - degree_2_loss: 0.6072 - quality_loss: 0.6218 - inversion_loss: 0.7708 - root_loss: 0.5484 - key_accura 5/12 [===========>..................] - ETA: 3s - loss: 3.2097 - key_loss: 0.3057 - degree_1_loss: 0.2972 - degree_2_loss: 0.6282 - quality_loss: 0.6410 - inversion_loss: 0.7806 - root_loss: 0.5570 - key_accura 6/12 [==============>...............] - ETA: 3s - loss: 3.2483 - key_loss: 0.3096 - degree_1_loss: 0.2943 - degree_2_loss: 0.6429 - quality_loss: 0.6550 - inversion_loss: 0.7825 - root_loss: 0.5641 - key_accura 7/12 [================>.............] - ETA: 2s - loss: 3.2875 - key_loss: 0.3146 - degree_1_loss: 0.2942 - degree_2_loss: 0.6555 - quality_loss: 0.6674 - inversion_loss: 0.7855 - root_loss: 0.5701 - key_accura 8/12 [===================>..........] - ETA: 2s - loss: 3.3233 - key_loss: 0.3190 - degree_1_loss: 0.2946 - degree_2_loss: 0.6670 - quality_loss: 0.6778 - inversion_loss: 0.7887 - root_loss: 0.5762 - key_accura 9/12 [=====================>........] - ETA: 1s - loss: 3.3593 - key_loss: 0.3237 - degree_1_loss: 0.2959 - degree_2_loss: 0.6779 - quality_loss: 0.6874 - inversion_loss: 0.7923 - root_loss: 0.5822 - key_accura10/12 [========================>.....] - ETA: 1s - loss: 3.3896 - key_loss: 0.3278 - degree_1_loss: 0.2969 - degree_2_loss: 0.6871 - quality_loss: 0.6951 - inversion_loss: 0.7960 - root_loss: 0.5867 - key_accura11/12 [==========================>...] - ETA: 0s - loss: 3.4146 - key_loss: 0.3311 - degree_1_loss: 0.2975 - degree_2_loss: 0.6947 - quality_loss: 0.7017 - inversion_loss: 0.7992 - root_loss: 0.5903 - key_accura12/12 [==============================] - ETA: 0s - loss: 3.4402 - key_loss: 0.3341 - degree_1_loss: 0.2982 - degree_2_loss: 0.7023 - quality_loss: 0.7082 - inversion_loss: 0.8022 - root_loss: 0.5952 - key_accura12/12 [==============================] - 7s 569ms/step - loss: 3.4620 - key_loss: 0.3366 - degree_1_loss: 0.2988 - degree_2_loss: 0.7088 - quality_loss: 0.7138 - inversion_loss: 0.8047 - root_loss: 0.5993 - key_accuracy: 0.9079 - degree_1_accuracy: 0.9108 - degree_2_accuracy: 0.7373 - quality_accuracy: 0.7253 - inversion_accuracy: 0.6302 - root_accuracy: 0.7933 - val_loss: 5.4845 - val_key_loss: 1.0519 - val_degree_1_loss: 0.3999 - val_degree_2_loss: 1.1355 - val_quality_loss: 1.0989 - val_inversion_loss: 0.9762 - val_root_loss: 0.8222 - val_key_accuracy: 0.6397 - val_degree_1_accuracy: 0.9072 - val_degree_2_accuracy: 0.5946 - val_quality_accuracy: 0.5799 - val_inversion_accuracy: 0.5921 - val_root_accuracy: 0.7454

loss: 3.4620
key_loss: 0.3366
degree_1_loss: 0.2988
degree_2_loss: 0.7088
quality_loss: 0.7138
inversion_loss: 0.8047
root_loss: 0.5993

key_accuracy: 0.9079
degree_1_accuracy: 0.9108
degree_2_accuracy: 0.7373
quality_accuracy: 0.7253
inversion_accuracy: 0.6302
root_accuracy: 0.7933

val_loss: 5.4845
val_key_loss: 1.0519
val_degree_1_loss: 0.3999
val_degree_2_loss: 1.1355
val_quality_loss: 1.0989
val_inversion_loss: 0.9762
val_root_loss: 0.8222

val_key_accuracy: 0.6397
val_degree_1_accuracy: 0.9072
val_degree_2_accuracy: 0.5946
val_quality_accuracy: 0.5799
val_inversion_accuracy: 0.5921
val_root_accuracy: 0.7454

Adding a CLI argument for controlling data_augmentation in a cleaner way: https://github.com/napulen/functional-harmony-micchi/commit/eddcf0f7f1933df65f9a05f0565448cb092d06ff

After using the new CLI interface for controlling data_augmentation

`python train.py --model 2 --input 4`

Epoch 75/100
 1/12 [=>............................] - ETA: 6s - loss: 4.0173 - key_loss: 0.4215 - degree_1_loss: 0.3732 - degree_2_loss: 0.8792 - quality_loss: 0.8696 - inversion_loss: 0.7812 - root_loss: 0.6926 - key_accura 2/12 [====>.........................] - ETA: 5s - loss: 3.9883 - key_loss: 0.4230 - degree_1_loss: 0.3686 - degree_2_loss: 0.8468 - quality_loss: 0.8518 - inversion_loss: 0.8095 - root_loss: 0.6886 - key_accura 3/12 [======>.......................] - ETA: 4s - loss: 3.9348 - key_loss: 0.4332 - degree_1_loss: 0.3526 - degree_2_loss: 0.8266 - quality_loss: 0.8316 - inversion_loss: 0.8177 - root_loss: 0.6731 - key_accura 4/12 [=========>....................] - ETA: 4s - loss: 3.9058 - key_loss: 0.4337 - degree_1_loss: 0.3403 - degree_2_loss: 0.8198 - quality_loss: 0.8189 - inversion_loss: 0.8225 - root_loss: 0.6706 - key_accura 5/12 [===========>..................] - ETA: 3s - loss: 3.9046 - key_loss: 0.4313 - degree_1_loss: 0.3348 - degree_2_loss: 0.8201 - quality_loss: 0.8150 - inversion_loss: 0.8300 - root_loss: 0.6734 - key_accura 6/12 [==============>...............] - ETA: 3s - loss: 3.9042 - key_loss: 0.4290 - degree_1_loss: 0.3322 - degree_2_loss: 0.8207 - quality_loss: 0.8130 - inversion_loss: 0.8336 - root_loss: 0.6756 - key_accura 7/12 [================>.............] - ETA: 2s - loss: 3.8941 - key_loss: 0.4249 - degree_1_loss: 0.3302 - degree_2_loss: 0.8189 - quality_loss: 0.8105 - inversion_loss: 0.8338 - root_loss: 0.6757 - key_accura 8/12 [===================>..........] - ETA: 2s - loss: 3.8882 - key_loss: 0.4229 - degree_1_loss: 0.3302 - degree_2_loss: 0.8168 - quality_loss: 0.8078 - inversion_loss: 0.8359 - root_loss: 0.6746 - key_accura 9/12 [=====================>........] - ETA: 1s - loss: 3.8820 - key_loss: 0.4213 - degree_1_loss: 0.3306 - degree_2_loss: 0.8146 - quality_loss: 0.8050 - inversion_loss: 0.8370 - root_loss: 0.6735 - key_accura10/12 [========================>.....] - ETA: 1s - loss: 3.8780 - key_loss: 0.4194 - degree_1_loss: 0.3309 - degree_2_loss: 0.8134 - quality_loss: 0.8031 - inversion_loss: 0.8381 - root_loss: 0.6732 - key_accura11/12 [==========================>...] - ETA: 0s - loss: 3.8776 - key_loss: 0.4185 - degree_1_loss: 0.3313 - degree_2_loss: 0.8130 - quality_loss: 0.8019 - inversion_loss: 0.8391 - root_loss: 0.6738 - key_accura12/12 [==============================] - ETA: 0s - loss: 3.8787 - key_loss: 0.4181 - degree_1_loss: 0.3314 - degree_2_loss: 0.8134 - quality_loss: 0.8012 - inversion_loss: 0.8403 - root_loss: 0.6743 - key_accura12/12 [==============================] - 7s 544ms/step - loss: 3.8797 - key_loss: 0.4177 - degree_1_loss: 0.3315 - degree_2_loss: 0.8137 - quality_loss: 0.8006 - inversion_loss: 0.8413 - root_loss: 0.6748 - key_accuracy: 0.8811 - degree_1_accuracy: 0.9093 - degree_2_accuracy: 0.7032 - quality_accuracy: 0.6970 - inversion_accuracy: 0.6231 - root_accuracy: 0.7684 - val_loss: 5.6914 - val_key_loss: 1.1656 - val_degree_1_loss: 0.4109 - val_degree_2_loss: 1.1968 - val_quality_loss: 1.1175 - val_inversion_loss: 0.9825 - val_root_loss: 0.8180 - val_key_accuracy: 0.6383 - val_degree_1_accuracy: 0.9045 - val_degree_2_accuracy: 0.5756 - val_quality_accuracy: 0.5880 - val_inversion_accuracy: 0.6043 - val_root_accuracy: 0.7440

loss: 3.8797
key_loss: 0.4177
degree_1_loss: 0.3315
degree_2_loss: 0.8137
quality_loss: 0.8006
inversion_loss: 0.8413
root_loss: 0.6748

key_accuracy: 0.8811
degree_1_accuracy: 0.9093
degree_2_accuracy: 0.7032
quality_accuracy: 0.6970
inversion_accuracy: 0.6231
root_accuracy: 0.7684

val_loss: 5.6914
val_key_loss: 1.1656
val_degree_1_loss: 0.4109
val_degree_2_loss: 1.1968
val_quality_loss: 1.1175
val_inversion_loss: 0.9825
val_root_loss: 0.8180

val_key_accuracy: 0.6383
val_degree_1_accuracy: 0.9045
val_degree_2_accuracy: 0.5756
val_quality_accuracy: 0.5880
val_inversion_accuracy: 0.6043
val_root_accuracy: 0.7440

Adding the code with my implementation of spelling_compressed_cut and some CLI/Config changes to everything else:

python train.py --model gru --input spelling_bass_cut

Epoch 72/100
 1/12 [=>............................] - ETA: 5s - loss: 4.2872 - key_loss: 0.4520 - degree_1_loss: 0.3513 - degree_2_loss: 0.9114 - quality_loss: 0.8880 - inversion_loss: 0.9291 - root_loss: 0.7555 - key_accura 2/12 [====>.........................] - ETA: 5s - loss: 4.1091 - key_loss: 0.4363 - degree_1_loss: 0.3478 - degree_2_loss: 0.8626 - quality_loss: 0.8548 - inversion_loss: 0.8945 - root_loss: 0.7131 - key_accura 3/12 [======>.......................] - ETA: 4s - loss: 4.0361 - key_loss: 0.4348 - degree_1_loss: 0.3395 - degree_2_loss: 0.8416 - quality_loss: 0.8383 - inversion_loss: 0.8871 - root_loss: 0.6948 - key_accura 4/12 [=========>....................] - ETA: 4s - loss: 3.9810 - key_loss: 0.4336 - degree_1_loss: 0.3365 - degree_2_loss: 0.8262 - quality_loss: 0.8258 - inversion_loss: 0.8772 - root_loss: 0.6817 - key_accura 5/12 [===========>..................] - ETA: 3s - loss: 3.9496 - key_loss: 0.4310 - degree_1_loss: 0.3380 - degree_2_loss: 0.8166 - quality_loss: 0.8180 - inversion_loss: 0.8723 - root_loss: 0.6736 - key_accura 6/12 [==============>...............] - ETA: 3s - loss: 3.9415 - key_loss: 0.4278 - degree_1_loss: 0.3385 - degree_2_loss: 0.8141 - quality_loss: 0.8161 - inversion_loss: 0.8712 - root_loss: 0.6738 - key_accura 7/12 [================>.............] - ETA: 2s - loss: 3.9319 - key_loss: 0.4243 - degree_1_loss: 0.3366 - degree_2_loss: 0.8125 - quality_loss: 0.8153 - inversion_loss: 0.8691 - root_loss: 0.6741 - key_accura 8/12 [===================>..........] - ETA: 2s - loss: 3.9316 - key_loss: 0.4228 - degree_1_loss: 0.3347 - degree_2_loss: 0.8123 - quality_loss: 0.8161 - inversion_loss: 0.8686 - root_loss: 0.6771 - key_accura 9/12 [=====================>........] - ETA: 1s - loss: 3.9227 - key_loss: 0.4198 - degree_1_loss: 0.3326 - degree_2_loss: 0.8109 - quality_loss: 0.8152 - inversion_loss: 0.8665 - root_loss: 0.6776 - key_accura10/12 [========================>.....] - ETA: 1s - loss: 3.9173 - key_loss: 0.4178 - degree_1_loss: 0.3307 - degree_2_loss: 0.8100 - quality_loss: 0.8152 - inversion_loss: 0.8648 - root_loss: 0.6788 - key_accura11/12 [==========================>...] - ETA: 0s - loss: 3.9153 - key_loss: 0.4164 - degree_1_loss: 0.3294 - degree_2_loss: 0.8101 - quality_loss: 0.8152 - inversion_loss: 0.8636 - root_loss: 0.6805 - key_accura12/12 [==============================] - ETA: 0s - loss: 3.9119 - key_loss: 0.4150 - degree_1_loss: 0.3284 - degree_2_loss: 0.8098 - quality_loss: 0.8146 - inversion_loss: 0.8628 - root_loss: 0.6812 - key_accura12/12 [==============================] - 7s 549ms/step - loss: 3.9090 - key_loss: 0.4139 - degree_1_loss: 0.3275 - degree_2_loss: 0.8095 - quality_loss: 0.8142 - inversion_loss: 0.8622 - root_loss: 0.6818 - key_accuracy: 0.8838 - degree_1_accuracy: 0.9035 - degree_2_accuracy: 0.7100 - quality_accuracy: 0.7064 - inversion_accuracy: 0.6250 - root_accuracy: 0.7733 - val_loss: 5.4991 - val_key_loss: 1.0954 - val_degree_1_loss: 0.3995 - val_degree_2_loss: 1.1559 - val_quality_loss: 1.0541 - val_inversion_loss: 0.9707 - val_root_loss: 0.8233 - val_key_accuracy: 0.6444 - val_degree_1_accuracy: 0.9068 - val_degree_2_accuracy: 0.5872 - val_quality_accuracy: 0.6063 - val_inversion_accuracy: 0.6192 - val_root_accuracy: 0.7386

loss: 3.9090
key_loss: 0.4139
degree_1_loss: 0.3275
degree_2_loss: 0.8095
quality_loss: 0.8142
inversion_loss: 0.8622
root_loss: 0.6818

key_accuracy: 0.8838
degree_1_accuracy: 0.9035
degree_2_accuracy: 0.7100
quality_accuracy: 0.7064
inversion_accuracy: 0.6250
root_accuracy: 0.7733

val_loss: 5.4991
val_key_loss: 1.0954
val_degree_1_loss: 0.3995
val_degree_2_loss: 1.1559
val_quality_loss: 1.0541
val_inversion_loss: 0.9707
val_root_loss: 0.8233

val_key_accuracy: 0.6444
val_degree_1_accuracy: 0.9068
val_degree_2_accuracy: 0.5872
val_quality_accuracy: 0.6063
val_inversion_accuracy: 0.6192
val_root_accuracy: 0.7386

It didn't seem to affect the results. Moving forward with my spelling_compressed_cut now.

First experiment with self-implemented spelling_compressed_cut. In this one, I am discarding octaves that exceed the range of (min_octave=2, max_octave=6).

            while octave < MIN_OCTAVE:
                logger.warning("Score lower than octave boundaries. Transposing up.")
                octave += 1
                continue
            while octave > MAX_OCTAVE:
                logger.warning("Score higher than octave boundaries. Transposing down.")
                octave -= 1
                continue

python train.py --model gru --input spelling_compressed_cut

Epoch 61/100
 1/12 [=>............................] - ETA: 5s - loss: 4.1116 - key_loss: 0.4852 - degree_1_loss: 0.3558 - degree_2_loss: 0.8805 - quality_loss: 0.8082 - inversion_loss: 0.8422 - root_loss: 0.7396 - key_accura 2/12 [====>.........................] - ETA: 5s - loss: 4.0431 - key_loss: 0.4675 - degree_1_loss: 0.3423 - degree_2_loss: 0.8736 - quality_loss: 0.8193 - inversion_loss: 0.8281 - root_loss: 0.7123 - key_accura 3/12 [======>.......................] - ETA: 4s - loss: 4.0085 - key_loss: 0.4602 - degree_1_loss: 0.3285 - degree_2_loss: 0.8669 - quality_loss: 0.8206 - inversion_loss: 0.8268 - root_loss: 0.7055 - key_accura 4/12 [=========>....................] - ETA: 4s - loss: 3.9978 - key_loss: 0.4622 - degree_1_loss: 0.3252 - degree_2_loss: 0.8645 - quality_loss: 0.8222 - inversion_loss: 0.8229 - root_loss: 0.7008 - key_accura 5/12 [===========>..................] - ETA: 3s - loss: 3.9986 - key_loss: 0.4675 - degree_1_loss: 0.3261 - degree_2_loss: 0.8640 - quality_loss: 0.8210 - inversion_loss: 0.8252 - root_loss: 0.6948 - key_accura 6/12 [==============>...............] - ETA: 3s - loss: 4.0194 - key_loss: 0.4773 - degree_1_loss: 0.3286 - degree_2_loss: 0.8663 - quality_loss: 0.8241 - inversion_loss: 0.8281 - root_loss: 0.6950 - key_accura 7/12 [================>.............] - ETA: 2s - loss: 4.0222 - key_loss: 0.4826 - degree_1_loss: 0.3285 - degree_2_loss: 0.8645 - quality_loss: 0.8251 - inversion_loss: 0.8282 - root_loss: 0.6934 - key_accura 8/12 [===================>..........] - ETA: 2s - loss: 4.0226 - key_loss: 0.4864 - degree_1_loss: 0.3294 - degree_2_loss: 0.8621 - quality_loss: 0.8257 - inversion_loss: 0.8278 - root_loss: 0.6912 - key_accura 9/12 [=====================>........] - ETA: 1s - loss: 4.0145 - key_loss: 0.4870 - degree_1_loss: 0.3290 - degree_2_loss: 0.8580 - quality_loss: 0.8247 - inversion_loss: 0.8276 - root_loss: 0.6881 - key_accura10/12 [========================>.....] - ETA: 1s - loss: 4.0134 - key_loss: 0.4875 - degree_1_loss: 0.3294 - degree_2_loss: 0.8562 - quality_loss: 0.8251 - inversion_loss: 0.8284 - root_loss: 0.6868 - key_accura11/12 [==========================>...] - ETA: 0s - loss: 4.0157 - key_loss: 0.4882 - degree_1_loss: 0.3296 - degree_2_loss: 0.8555 - quality_loss: 0.8261 - inversion_loss: 0.8301 - root_loss: 0.6862 - key_accura12/12 [==============================] - ETA: 0s - loss: 4.0174 - key_loss: 0.4878 - degree_1_loss: 0.3302 - degree_2_loss: 0.8551 - quality_loss: 0.8269 - inversion_loss: 0.8318 - root_loss: 0.6857 - key_accura12/12 [==============================] - 7s 566ms/step - loss: 4.0189 - key_loss: 0.4874 - degree_1_loss: 0.3306 - degree_2_loss: 0.8548 - quality_loss: 0.8276 - inversion_loss: 0.8333 - root_loss: 0.6852 - key_accuracy: 0.8682 - degree_1_accuracy: 0.9101 - degree_2_accuracy: 0.6937 - quality_accuracy: 0.6996 - inversion_accuracy: 0.6407 - root_accuracy: 0.7739 - val_loss: 5.7103 - val_key_loss: 1.1787 - val_degree_1_loss: 0.3974 - val_degree_2_loss: 1.2088 - val_quality_loss: 1.1139 - val_inversion_loss: 1.0020 - val_root_loss: 0.8094 - val_key_accuracy: 0.6190 - val_degree_1_accuracy: 0.9080 - val_degree_2_accuracy: 0.5674 - val_quality_accuracy: 0.5733 - val_inversion_accuracy: 0.5964 - val_root_accuracy: 0.7449

loss: 4.0189
key_loss: 0.4874
degree_1_loss: 0.3306
degree_2_loss: 0.8548
quality_loss: 0.8276
inversion_loss: 0.8333
root_loss: 0.6852

key_accuracy: 0.8682
degree_1_accuracy: 0.9101
degree_2_accuracy: 0.6937
quality_accuracy: 0.6996
inversion_accuracy: 0.6407
root_accuracy: 0.7739

val_loss: 5.7103
val_key_loss: 1.1787
val_degree_1_loss: 0.3974
val_degree_2_loss: 1.2088
val_quality_loss: 1.1139
val_inversion_loss: 1.0020
val_root_loss: 0.8094

val_key_accuracy: 0.6190
val_degree_1_accuracy: 0.9080
val_degree_2_accuracy: 0.5674
val_quality_accuracy: 0.5733
val_inversion_accuracy: 0.5964
val_root_accuracy: 0.7449

The performance is a bit lower than spelling_bass_cut, but I guess there's potential.

Originally, what I wanted to do is see if sending all notes below/above the limits to the lowest/highest octave (to not entire lose the information) would be better. That's the next experiment.

python train.py --model gru --input spelling_compressed_cut

Epoch 63/100
 1/12 [=>............................] - ETA: 5s - loss: 4.2736 - key_loss: 0.5001 - degree_1_loss: 0.3670 - degree_2_loss: 0.9540 - quality_loss: 0.8968 - inversion_loss: 0.8822 - root_loss: 0.6735 - key_accura 2/12 [====>.........................] - ETA: 5s - loss: 4.2002 - key_loss: 0.5022 - degree_1_loss: 0.3471 - degree_2_loss: 0.9378 - quality_loss: 0.8856 - inversion_loss: 0.8647 - root_loss: 0.6629 - key_accura 3/12 [======>.......................] - ETA: 4s - loss: 4.1500 - key_loss: 0.4879 - degree_1_loss: 0.3399 - degree_2_loss: 0.9272 - quality_loss: 0.8793 - inversion_loss: 0.8540 - root_loss: 0.6617 - key_accura 4/12 [=========>....................] - ETA: 4s - loss: 4.1502 - key_loss: 0.4830 - degree_1_loss: 0.3362 - degree_2_loss: 0.9256 - quality_loss: 0.8808 - inversion_loss: 0.8571 - root_loss: 0.6676 - key_accura 5/12 [===========>..................] - ETA: 3s - loss: 4.1518 - key_loss: 0.4813 - degree_1_loss: 0.3382 - degree_2_loss: 0.9218 - quality_loss: 0.8783 - inversion_loss: 0.8603 - root_loss: 0.6719 - key_accura 6/12 [==============>...............] - ETA: 3s - loss: 4.1317 - key_loss: 0.4774 - degree_1_loss: 0.3371 - degree_2_loss: 0.9138 - quality_loss: 0.8711 - inversion_loss: 0.8615 - root_loss: 0.6707 - key_accura 7/12 [================>.............] - ETA: 2s - loss: 4.1228 - key_loss: 0.4751 - degree_1_loss: 0.3375 - degree_2_loss: 0.9084 - quality_loss: 0.8671 - inversion_loss: 0.8642 - root_loss: 0.6704 - key_accura 8/12 [===================>..........] - ETA: 2s - loss: 4.1110 - key_loss: 0.4730 - degree_1_loss: 0.3377 - degree_2_loss: 0.9028 - quality_loss: 0.8622 - inversion_loss: 0.8663 - root_loss: 0.6690 - key_accura 9/12 [=====================>........] - ETA: 1s - loss: 4.1029 - key_loss: 0.4710 - degree_1_loss: 0.3376 - degree_2_loss: 0.8991 - quality_loss: 0.8591 - inversion_loss: 0.8674 - root_loss: 0.6687 - key_accura10/12 [========================>.....] - ETA: 1s - loss: 4.0911 - key_loss: 0.4688 - degree_1_loss: 0.3375 - degree_2_loss: 0.8944 - quality_loss: 0.8554 - inversion_loss: 0.8674 - root_loss: 0.6676 - key_accura11/12 [==========================>...] - ETA: 0s - loss: 4.0808 - key_loss: 0.4673 - degree_1_loss: 0.3373 - degree_2_loss: 0.8903 - quality_loss: 0.8523 - inversion_loss: 0.8671 - root_loss: 0.6666 - key_accura12/12 [==============================] - ETA: 0s - loss: 4.0737 - key_loss: 0.4662 - degree_1_loss: 0.3373 - degree_2_loss: 0.8871 - quality_loss: 0.8502 - inversion_loss: 0.8667 - root_loss: 0.6661 - key_accura12/12 [==============================] - 7s 555ms/step - loss: 4.0676 - key_loss: 0.4654 - degree_1_loss: 0.3372 - degree_2_loss: 0.8844 - quality_loss: 0.8484 - inversion_loss: 0.8665 - root_loss: 0.6657 - key_accuracy: 0.8759 - degree_1_accuracy: 0.9100 - degree_2_accuracy: 0.6851 - quality_accuracy: 0.6950 - inversion_accuracy: 0.6202 - root_accuracy: 0.7813 - val_loss: 5.7301 - val_key_loss: 1.1769 - val_degree_1_loss: 0.4143 - val_degree_2_loss: 1.1744 - val_quality_loss: 1.1420 - val_inversion_loss: 0.9958 - val_root_loss: 0.8267 - val_key_accuracy: 0.6176 - val_degree_1_accuracy: 0.9099 - val_degree_2_accuracy: 0.5918 - val_quality_accuracy: 0.5574 - val_inversion_accuracy: 0.5975 - val_root_accuracy: 0.7352

loss: 4.0676
key_loss: 0.4654
degree_1_loss: 0.3372
degree_2_loss: 0.8844
quality_loss: 0.8484
inversion_loss: 0.8665
root_loss: 0.6657

key_accuracy: 0.8759
degree_1_accuracy: 0.9100
degree_2_accuracy: 0.6851
quality_accuracy: 0.6950
inversion_accuracy: 0.6202
root_accuracy: 0.7813

val_loss: 5.7301
val_key_loss: 1.1769
val_degree_1_loss: 0.4143
val_degree_2_loss: 1.1744
val_quality_loss: 1.1420
val_inversion_loss: 0.9958
val_root_loss: 0.8267

val_key_accuracy: 0.6176
val_degree_1_accuracy: 0.9099
val_degree_2_accuracy: 0.5918
val_quality_accuracy: 0.5574
val_inversion_accuracy: 0.5975
val_root_accuracy: 0.7352

An idea for a representation that compresses spelling into 19 features instead of 35: A chromagram plus the diatonic classes (white keys) that are sounding.

In similar experiments to those I have been running with spelling_bass_cut and spelling_compressed_cut, it did pretty well:

Epoch 80/100
 1/13 [=>............................] - ETA: 6s - loss: 3.0435 - key_loss: 0.2482 - degree_1_loss: 0.2632 - degree_2_loss: 0.6024 - quality_loss: 0.5963 - inversion_loss: 0.7618 - root_loss: 0.5716 - key_accura 2/13 [===>..........................] - ETA: 5s - loss: 3.2081 - key_loss: 0.3015 - degree_1_loss: 0.2934 - degree_2_loss: 0.6305 - quality_loss: 0.6254 - inversion_loss: 0.7719 - root_loss: 0.5853 - key_accura 3/13 [=====>........................] - ETA: 5s - loss: 3.3576 - key_loss: 0.3351 - degree_1_loss: 0.3016 - degree_2_loss: 0.6592 - quality_loss: 0.6607 - inversion_loss: 0.7902 - root_loss: 0.6108 - key_accura 4/13 [========>.....................] - ETA: 4s - loss: 3.4612 - key_loss: 0.3569 - degree_1_loss: 0.3077 - degree_2_loss: 0.6788 - quality_loss: 0.6875 - inversion_loss: 0.8042 - root_loss: 0.6262 - key_accura 5/13 [==========>...................] - ETA: 4s - loss: 3.5335 - key_loss: 0.3644 - degree_1_loss: 0.3099 - degree_2_loss: 0.6971 - quality_loss: 0.7081 - inversion_loss: 0.8145 - root_loss: 0.6396 - key_accura 6/13 [============>.................] - ETA: 3s - loss: 3.5791 - key_loss: 0.3676 - degree_1_loss: 0.3097 - degree_2_loss: 0.7087 - quality_loss: 0.7229 - inversion_loss: 0.8206 - root_loss: 0.6496 - key_accura 7/13 [===============>..............] - ETA: 3s - loss: 3.6138 - key_loss: 0.3698 - degree_1_loss: 0.3093 - degree_2_loss: 0.7189 - quality_loss: 0.7336 - inversion_loss: 0.8262 - root_loss: 0.6560 - key_accura 8/13 [=================>............] - ETA: 2s - loss: 3.6464 - key_loss: 0.3731 - degree_1_loss: 0.3102 - degree_2_loss: 0.7286 - quality_loss: 0.7439 - inversion_loss: 0.8297 - root_loss: 0.6610 - key_accura 9/13 [===================>..........] - ETA: 2s - loss: 3.6726 - key_loss: 0.3764 - degree_1_loss: 0.3112 - degree_2_loss: 0.7361 - quality_loss: 0.7517 - inversion_loss: 0.8328 - root_loss: 0.6645 - key_accura10/13 [======================>.......] - ETA: 1s - loss: 3.6942 - key_loss: 0.3792 - degree_1_loss: 0.3122 - degree_2_loss: 0.7420 - quality_loss: 0.7578 - inversion_loss: 0.8355 - root_loss: 0.6675 - key_accura11/13 [========================>.....] - ETA: 1s - loss: 3.7163 - key_loss: 0.3822 - degree_1_loss: 0.3135 - degree_2_loss: 0.7475 - quality_loss: 0.7639 - inversion_loss: 0.8383 - root_loss: 0.6709 - key_accura12/13 [==========================>...] - ETA: 0s - loss: 3.7336 - key_loss: 0.3840 - degree_1_loss: 0.3142 - degree_2_loss: 0.7522 - quality_loss: 0.7690 - inversion_loss: 0.8403 - root_loss: 0.6739 - key_accura13/13 [==============================] - ETA: 0s - loss: 3.7493 - key_loss: 0.3856 - degree_1_loss: 0.3148 - degree_2_loss: 0.7563 - quality_loss: 0.7735 - inversion_loss: 0.8422 - root_loss: 0.6769 - key_accura13/13 [==============================] - 7s 549ms/step - loss: 3.7627 - key_loss: 0.3869 - degree_1_loss: 0.3153 - degree_2_loss: 0.7598 - quality_loss: 0.7774 - inversion_loss: 0.8439 - root_loss: 0.6794 - key_accuracy: 0.8844 - degree_1_accuracy: 0.9094 - degree_2_accuracy: 0.7267 - quality_accuracy: 0.7107 - inversion_accuracy: 0.6253 - root_accuracy: 0.7697 - val_loss: 5.5509 - val_key_loss: 1.1527 - val_degree_1_loss: 0.3908 - val_degree_2_loss: 1.1096 - val_quality_loss: 1.0607 - val_inversion_loss: 0.9942 - val_root_loss: 0.8428 - val_key_accuracy: 0.6599 - val_degree_1_accuracy: 0.9079 - val_degree_2_accuracy: 0.5913 - val_quality_accuracy: 0.5936 - val_inversion_accuracy: 0.5984 - val_root_accuracy: 0.7362

loss: 3.7627
key_loss: 0.3869
degree_1_loss: 0.3153
degree_2_loss: 0.7598
quality_loss: 0.7774
inversion_loss: 0.8439
root_loss: 0.6794

key_accuracy: 0.8844
degree_1_accuracy: 0.9094
degree_2_accuracy: 0.7267
quality_accuracy: 0.7107
inversion_accuracy: 0.6253
root_accuracy: 0.7697

val_loss: 5.5509
val_key_loss: 1.1527
val_degree_1_loss: 0.3908
val_degree_2_loss: 1.1096
val_quality_loss: 1.0607
val_inversion_loss: 0.9942
val_root_loss: 0.8428

val_key_accuracy: 0.6599
val_degree_1_accuracy: 0.9079
val_degree_2_accuracy: 0.5913
val_quality_accuracy: 0.5936
val_inversion_accuracy: 0.5984
val_root_accuracy: 0.7362

I'm liking this representation. Possible paths to pursue:

Encode multiple octaves by encoding a 7*octaves vector and a global 12 chromagram vector
Encode multiple octaves by encoding a (7+12)*octaves vector. This is equivalent to the full-spelled representation Micchi et al. has with 245 parameters. This would only take 133 parameters to cover the full 7-octave range, theoretically, same information. I would prefer 5 octave over 7, with transposition of the lower/higher octaves to lowest/highest octave of the covered range.

As a first experiment, here is one trial with 12 + (7*3). The 3-octave range is C3 to B5, transposing anything outside the range to the lowest-highest octave of the range.

Epoch 65/100
 1/13 [=>............................] - ETA: 6s - loss: 4.2205 - key_loss: 0.3900 - degree_1_loss: 0.3793 - degree_2_loss: 0.9191 - quality_loss: 0.8893 - inversion_loss: 0.8637 - root_loss: 0.7792 - key_accura 2/13 [===>..........................] - ETA: 5s - loss: 4.2433 - key_loss: 0.3937 - degree_1_loss: 0.3891 - degree_2_loss: 0.9131 - quality_loss: 0.8775 - inversion_loss: 0.8767 - root_loss: 0.7933 - key_accura 3/13 [=====>........................] - ETA: 5s - loss: 4.2087 - key_loss: 0.4032 - degree_1_loss: 0.3820 - degree_2_loss: 0.8984 - quality_loss: 0.8601 - inversion_loss: 0.8743 - root_loss: 0.7906 - key_accura 4/13 [========>.....................] - ETA: 4s - loss: 4.2016 - key_loss: 0.4131 - degree_1_loss: 0.3756 - degree_2_loss: 0.8912 - quality_loss: 0.8558 - inversion_loss: 0.8780 - root_loss: 0.7880 - key_accura 5/13 [==========>...................] - ETA: 4s - loss: 4.2042 - key_loss: 0.4151 - degree_1_loss: 0.3726 - degree_2_loss: 0.8890 - quality_loss: 0.8569 - inversion_loss: 0.8821 - root_loss: 0.7885 - key_accura 6/13 [============>.................] - ETA: 3s - loss: 4.1885 - key_loss: 0.4176 - degree_1_loss: 0.3675 - degree_2_loss: 0.8835 - quality_loss: 0.8561 - inversion_loss: 0.8803 - root_loss: 0.7835 - key_accura 7/13 [===============>..............] - ETA: 3s - loss: 4.1579 - key_loss: 0.4164 - degree_1_loss: 0.3633 - degree_2_loss: 0.8759 - quality_loss: 0.8517 - inversion_loss: 0.8760 - root_loss: 0.7747 - key_accura 8/13 [=================>............] - ETA: 2s - loss: 4.1338 - key_loss: 0.4145 - degree_1_loss: 0.3605 - degree_2_loss: 0.8702 - quality_loss: 0.8480 - inversion_loss: 0.8725 - root_loss: 0.7681 - key_accura 9/13 [===================>..........] - ETA: 2s - loss: 4.1109 - key_loss: 0.4117 - degree_1_loss: 0.3575 - degree_2_loss: 0.8646 - quality_loss: 0.8442 - inversion_loss: 0.8710 - root_loss: 0.7619 - key_accura10/13 [======================>.......] - ETA: 1s - loss: 4.0920 - key_loss: 0.4099 - degree_1_loss: 0.3544 - degree_2_loss: 0.8607 - quality_loss: 0.8412 - inversion_loss: 0.8690 - root_loss: 0.7568 - key_accura11/13 [========================>.....] - ETA: 1s - loss: 4.0823 - key_loss: 0.4090 - degree_1_loss: 0.3523 - degree_2_loss: 0.8586 - quality_loss: 0.8401 - inversion_loss: 0.8685 - root_loss: 0.7538 - key_accura12/13 [==========================>...] - ETA: 0s - loss: 4.0781 - key_loss: 0.4084 - degree_1_loss: 0.3508 - degree_2_loss: 0.8577 - quality_loss: 0.8400 - inversion_loss: 0.8691 - root_loss: 0.7522 - key_accura13/13 [==============================] - ETA: 0s - loss: 4.0745 - key_loss: 0.4079 - degree_1_loss: 0.3495 - degree_2_loss: 0.8569 - quality_loss: 0.8401 - inversion_loss: 0.8692 - root_loss: 0.7508 - key_accura13/13 [==============================] - 7s 542ms/step - loss: 4.0713 - key_loss: 0.4076 - degree_1_loss: 0.3484 - degree_2_loss: 0.8563 - quality_loss: 0.8402 - inversion_loss: 0.8693 - root_loss: 0.7496 - key_accuracy: 0.8856 - degree_1_accuracy: 0.9041 - degree_2_accuracy: 0.6901 - quality_accuracy: 0.6951 - inversion_accuracy: 0.6198 - root_accuracy: 0.7577 - val_loss: 5.5771 - val_key_loss: 1.0883 - val_degree_1_loss: 0.4071 - val_degree_2_loss: 1.1331 - val_quality_loss: 1.0928 - val_inversion_loss: 1.0017 - val_root_loss: 0.8541 - val_key_accuracy: 0.6584 - val_degree_1_accuracy: 0.9081 - val_degree_2_accuracy: 0.5858 - val_quality_accuracy: 0.5824 - val_inversion_accuracy: 0.6057 - val_root_accuracy: 0.7242

key_loss: 0.4076
degree_1_loss: 0.3484
degree_2_loss: 0.8563
quality_loss: 0.8402
inversion_loss: 0.8693
root_loss: 0.7496

key_accuracy: 0.8856
degree_1_accuracy: 0.9041
degree_2_accuracy: 0.6901
quality_accuracy: 0.6951
inversion_accuracy: 0.6198
root_accuracy: 0.7577

val_loss: 5.5771
val_key_loss: 1.0883
val_degree_1_loss: 0.4071
val_degree_2_loss: 1.1331
val_quality_loss: 1.0928
val_inversion_loss: 1.0017
val_root_loss: 0.8541

val_key_accuracy: 0.6584
val_degree_1_accuracy: 0.9081
val_degree_2_accuracy: 0.5858
val_quality_accuracy: 0.5824
val_inversion_accuracy: 0.6057
val_root_accuracy: 0.7242

Not doing particularly better. It's super weird that no model with actually more information about the bass has been able to provide accurate inversion predictions, but the 19 hybrid chromagram outperforms them by a little bit, with no information about the bass (or inversion) whatsoever. Something buggy with the code there I think.

Last thing I wanna try with this one before moving on adding augmentation: having the (12+7)*octave with 3 octaves. I'd like to see if per-octave chromagrams outperform the global chromagram that has less parameters.

Epoch 64/100
 1/13 [=>............................] - ETA: 6s - loss: 3.4551 - key_loss: 0.4248 - degree_1_loss: 0.2688 - degree_2_loss: 0.6653 - quality_loss: 0.6869 - inversion_loss: 0.8128 - root_loss: 0.5967 - key_accura 2/13 [===>..........................] - ETA: 5s - loss: 3.6043 - key_loss: 0.4170 - degree_1_loss: 0.3054 - degree_2_loss: 0.7136 - quality_loss: 0.7199 - inversion_loss: 0.8163 - root_loss: 0.6320 - key_accura 3/13 [=====>........................] - ETA: 4s - loss: 3.6301 - key_loss: 0.4036 - degree_1_loss: 0.3085 - degree_2_loss: 0.7271 - quality_loss: 0.7378 - inversion_loss: 0.8157 - root_loss: 0.6373 - key_accura 4/13 [========>.....................] - ETA: 4s - loss: 3.6254 - key_loss: 0.3940 - degree_1_loss: 0.3037 - degree_2_loss: 0.7321 - quality_loss: 0.7468 - inversion_loss: 0.8077 - root_loss: 0.6411 - key_accura 5/13 [==========>...................] - ETA: 3s - loss: 3.6372 - key_loss: 0.3890 - degree_1_loss: 0.3017 - degree_2_loss: 0.7404 - quality_loss: 0.7567 - inversion_loss: 0.8016 - root_loss: 0.6478 - key_accura 6/13 [============>.................] - ETA: 3s - loss: 3.6592 - key_loss: 0.3866 - degree_1_loss: 0.3016 - degree_2_loss: 0.7486 - quality_loss: 0.7671 - inversion_loss: 0.7989 - root_loss: 0.6564 - key_accura 7/13 [===============>..............] - ETA: 3s - loss: 3.6901 - key_loss: 0.3873 - degree_1_loss: 0.3048 - degree_2_loss: 0.7576 - quality_loss: 0.7763 - inversion_loss: 0.7978 - root_loss: 0.6663 - key_accura 8/13 [=================>............] - ETA: 2s - loss: 3.7140 - key_loss: 0.3889 - degree_1_loss: 0.3066 - degree_2_loss: 0.7645 - quality_loss: 0.7836 - inversion_loss: 0.7968 - root_loss: 0.6737 - key_accura 9/13 [===================>..........] - ETA: 1s - loss: 3.7346 - key_loss: 0.3897 - degree_1_loss: 0.3087 - degree_2_loss: 0.7699 - quality_loss: 0.7891 - inversion_loss: 0.7970 - root_loss: 0.6803 - key_accura10/13 [======================>.......] - ETA: 1s - loss: 3.7520 - key_loss: 0.3898 - degree_1_loss: 0.3104 - degree_2_loss: 0.7746 - quality_loss: 0.7936 - inversion_loss: 0.7974 - root_loss: 0.6862 - key_accura11/13 [========================>.....] - ETA: 1s - loss: 3.7652 - key_loss: 0.3895 - degree_1_loss: 0.3115 - degree_2_loss: 0.7781 - quality_loss: 0.7972 - inversion_loss: 0.7977 - root_loss: 0.6912 - key_accura12/13 [==========================>...] - ETA: 0s - loss: 3.7740 - key_loss: 0.3889 - degree_1_loss: 0.3123 - degree_2_loss: 0.7807 - quality_loss: 0.7996 - inversion_loss: 0.7979 - root_loss: 0.6945 - key_accura13/13 [==============================] - ETA: 0s - loss: 3.7819 - key_loss: 0.3886 - degree_1_loss: 0.3129 - degree_2_loss: 0.7829 - quality_loss: 0.8017 - inversion_loss: 0.7984 - root_loss: 0.6975 - key_accura13/13 [==============================] - 7s 540ms/step - loss: 3.7887 - key_loss: 0.3882 - degree_1_loss: 0.3135 - degree_2_loss: 0.7848 - quality_loss: 0.8034 - inversion_loss: 0.7987 - root_loss: 0.7000 - key_accuracy: 0.8864 - degree_1_accuracy: 0.9127 - degree_2_accuracy: 0.6992 - quality_accuracy: 0.6866 - inversion_accuracy: 0.6274 - root_accuracy: 0.7494 - val_loss: 5.3009 - val_key_loss: 1.0596 - val_degree_1_loss: 0.3703 - val_degree_2_loss: 1.0646 - val_quality_loss: 1.0488 - val_inversion_loss: 0.9011 - val_root_loss: 0.8565 - val_key_accuracy: 0.6355 - val_degree_1_accuracy: 0.9072 - val_degree_2_accuracy: 0.5610 - val_quality_accuracy: 0.5399 - val_inversion_accuracy: 0.6041 - val_root_accuracy: 0.6860

loss: 3.7887
key_loss: 0.3882
degree_1_loss: 0.3135
degree_2_loss: 0.7848
quality_loss: 0.8034
inversion_loss: 0.7987
root_loss: 0.7000

key_accuracy: 0.8864
degree_1_accuracy: 0.9127
degree_2_accuracy: 0.6992
quality_accuracy: 0.6866
inversion_accuracy: 0.6274
root_accuracy: 0.7494

val_loss: 5.3009
val_key_loss: 1.0596
val_degree_1_loss: 0.3703
val_degree_2_loss: 1.0646
val_quality_loss: 1.0488
val_inversion_loss: 0.9011
val_root_loss: 0.8565

val_key_accuracy: 0.6355
val_degree_1_accuracy: 0.9072
val_degree_2_accuracy: 0.5610
val_quality_accuracy: 0.5399
val_inversion_accuracy: 0.6041
val_root_accuracy: 0.6860

It didn't do better. It didn't do worse either. More data will tell.

For now, let's implement data augmentation (transposition) with this pitch_hybrid_cut representation. Then re-run the comparison against spelling_bass_cut with conv_gru and data augmentation.

Something really important I missed last time:

The evaluation of pitch and spelling models is different. The fact that pitch_hybrid_cut was better last time, doesn't necessarily mean it's better because the task is easier for that representation than it is for spelling_bass_cut

Therefore, today I am running two experiments with pitch_bass_cut and pitch_hybrid_cut against each other. That should really tell the difference.

Augmentation is on now (which I recently implemented for pitch_hybrid_cut.

python train.py --model gru --input pitch_bass_cut

Epoch 23/100
  1/150 [..............................] - ETA: 1:25 - loss: 3.8159 - key_loss: 0.4343 - degree_1_loss: 0.2958 - degree_2_loss: 0.7101 - quality_loss: 0.7406 - inversion_loss: 0.9076 - root_loss: 0.7274 - key_ac  2/150 [..............................] - ETA: 1:16 - loss: 3.8291 - key_loss: 0.4045 - degree_1_loss: 0.3520 - degree_2_loss: 0.7240 - quality_loss: 0.7282 - inversion_loss: 0.9344 - root_loss: 0.6858 - key_ac  3/150 [..............................] - ETA: 1:15 - loss: 3.8567 - key_loss: 0.4200 - degree_1_loss: 0.3682 - degree_2_loss: 0.7271 - quality_loss: 0.7316 - inversion_loss: 0.9308 - root_loss: 0.6790 - key_ac  4/150 [..............................] - ETA: 1:14 - loss: 3.9023 - key_loss: 0.4327 - degree_1_loss: 0.3721 - degree_2_loss: 0.7393 - quality_loss: 0.7437 - inversion_loss: 0.9304 - root_loss: 0.6842 - key_ac  5/150 [>.............................] - ETA: 1:14 - loss: 3.9357 - key_loss: 0.4421 - degree_1_loss: 0.3771 - degree_2_loss: 0.7489 - quality_loss: 0.7507 - inversion_loss: 0.9293 - root_loss: 0.6876 - key_ac  6/150 [>.............................] - ETA: 1:13 - loss: 3.9651 - key_loss: 0.4527 - degree_1_loss: 0.3781 - degree_2_loss: 0.7562 - quality_loss: 0.7573 - inversion_loss: 0.9267 - root_loss: 0.6941 - key_ac  7/150 [>.............................] - ETA: 1:13 - loss: 3.9963 - key_loss: 0.4619 - degree_1_loss: 0.3803 - degree_2_loss: 0.7628 - quality_loss: 0.7633 - inversion_loss: 0.9254 - root_loss: 0.7025 - key_ac  8/150 [>.............................] - ETA: 1:12 - loss: 4.0144 - key_loss: 0.4671 - degree_1_loss: 0.3805 - degree_2_loss: 0.7677 - quality_loss: 0.7673 - inversion_loss: 0.9233 - root_loss: 0.7085 - key_ac  9/150 [>.............................] - ETA: 1:12 - loss: 4.0190 - key_loss: 0.4690 - degree_1_loss: 0.3792 - degree_2_loss: 0.7704 - quality_loss: 0.7687 - inversion_loss: 0.9206 - root_loss: 0.7109 - key_ac 10/150 [=>............................] - ETA: 1:11 - loss: 4.0193 - key_loss: 0.4691 - degree_1_loss: 0.3777 - degree_2_loss: 0.7723 - quality_loss: 0.7693 - inversion_loss: 0.9191 - root_loss: 0.7117 - key_ac 11/150 [=>............................] - ETA: 1:11 - loss: 4.0185 - key_loss: 0.4695 - degree_1_loss: 0.3767 - degree_2_loss: 0.7741 - quality_loss: 0.7693 - inversion_loss: 0.9174 - root_loss: 0.7115 - key_ac 12/150 [=>............................] - ETA: 1:10 - loss: 4.0179 - key_loss: 0.4696 - degree_1_loss: 0.3759 - degree_2_loss: 0.7761 - quality_loss: 0.7694 - inversion_loss: 0.9156 - root_loss: 0.7114 - key_ac 13/150 [=>............................] - ETA: 1:10 - loss: 4.0186 - key_loss: 0.4700 - degree_1_loss: 0.3751 - degree_2_loss: 0.7780 - quality_loss: 0.7697 - inversion_loss: 0.9142 - root_loss: 0.7117 - key_ac 14/150 [=>............................] - ETA: 1:09 - loss: 4.0212 - key_loss: 0.4714 - degree_1_loss: 0.3751 - degree_2_loss: 0.7799 - quality_loss: 0.7698 - inversion_loss: 0.9129 - root_loss: 0.7121 - key_ac 15/150 [==>...........................] - ETA: 1:09 - loss: 4.0231 - key_loss: 0.4725 - degree_1_loss: 0.3750 - degree_2_loss: 0.7814 - quality_loss: 0.7700 - inversion_loss: 0.9117 - root_loss: 0.7125 - key_ac 16/150 [==>...........................] - ETA: 1:08 - loss: 4.0228 - key_loss: 0.4731 - degree_1_loss: 0.3747 - degree_2_loss: 0.7822 - quality_loss: 0.7700 - inversion_loss: 0.9103 - root_loss: 0.7124 - key_ac 17/150 [==>...........................] - ETA: 1:07 - loss: 4.0230 - key_loss: 0.4735 - degree_1_loss: 0.3744 - degree_2_loss: 0.7830 - quality_loss: 0.7703 - inversion_loss: 0.9091 - root_loss: 0.7126 - key_ac 18/150 [==>...........................] - ETA: 1:07 - loss: 4.0255 - key_loss: 0.4747 - degree_1_loss: 0.3743 - degree_2_loss: 0.7844 - quality_loss: 0.7710 - inversion_loss: 0.9081 - root_loss: 0.7130 - key_ac 19/150 [==>...........................] - ETA: 1:07 - loss: 4.0267 - key_loss: 0.4754 - degree_1_loss: 0.3740 - degree_2_loss: 0.7853 - quality_loss: 0.7716 - inversion_loss: 0.9072 - root_loss: 0.7132 - key_ac 20/150 [===>..........................] - ETA: 1:06 - loss: 4.0259 - key_loss: 0.4756 - degree_1_loss: 0.3734 - degree_2_loss: 0.7858 - quality_loss: 0.7718 - inversion_loss: 0.9061 - root_loss: 0.7131 - key_ac 21/150 [===>..........................] - ETA: 1:05 - loss: 4.0256 - key_loss: 0.4761 - degree_1_loss: 0.3728 - degree_2_loss: 0.7864 - quality_loss: 0.7719 - inversion_loss: 0.9052 - root_loss: 0.7131 - key_ac 22/150 [===>..........................] - ETA: 1:05 - loss: 4.0250 - key_loss: 0.4767 - degree_1_loss: 0.3723 - degree_2_loss: 0.7868 - quality_loss: 0.7720 - inversion_loss: 0.9043 - root_loss: 0.7130 - key_ac 23/150 [===>..........................] - ETA: 1:04 - loss: 4.0242 - key_loss: 0.4771 - degree_1_loss: 0.3718 - degree_2_loss: 0.7870 - quality_loss: 0.7720 - inversion_loss: 0.9036 - root_loss: 0.7128 - key_ac 24/150 [===>..........................] - ETA: 1:04 - loss: 4.0236 - key_loss: 0.4776 - degree_1_loss: 0.3714 - degree_2_loss: 0.7871 - quality_loss: 0.7719 - inversion_loss: 0.9031 - root_loss: 0.7125 - key_ac 25/150 [====>.........................] - ETA: 1:03 - loss: 4.0230 - key_loss: 0.4780 - degree_1_loss: 0.3709 - degree_2_loss: 0.7874 - quality_loss: 0.7718 - inversion_loss: 0.9027 - root_loss: 0.7122 - key_ac 26/150 [====>.........................] - ETA: 1:03 - loss: 4.0234 - key_loss: 0.4785 - degree_1_loss: 0.3706 - degree_2_loss: 0.7878 - quality_loss: 0.7720 - inversion_loss: 0.9024 - root_loss: 0.7121 - key_ac 27/150 [====>.........................] - ETA: 1:02 - loss: 4.0237 - key_loss: 0.4788 - degree_1_loss: 0.3703 - degree_2_loss: 0.7882 - quality_loss: 0.7722 - inversion_loss: 0.9021 - root_loss: 0.7120 - key_ac 28/150 [====>.........................] - ETA: 1:02 - loss: 4.0239 - key_loss: 0.4791 - degree_1_loss: 0.3699 - degree_2_loss: 0.7886 - quality_loss: 0.7724 - inversion_loss: 0.9019 - root_loss: 0.7120 - key_ac 29/150 [====>.........................] - ETA: 1:01 - loss: 4.0243 - key_loss: 0.4792 - degree_1_loss: 0.3696 - degree_2_loss: 0.7890 - quality_loss: 0.7726 - inversion_loss: 0.9019 - root_loss: 0.7119 - key_ac 30/150 [=====>........................] - ETA: 1:01 - loss: 4.0236 - key_loss: 0.4792 - degree_1_loss: 0.3693 - degree_2_loss: 0.7891 - quality_loss: 0.7726 - inversion_loss: 0.9018 - root_loss: 0.7117 - key_ac 31/150 [=====>........................] - ETA: 1:00 - loss: 4.0235 - key_loss: 0.4792 - degree_1_loss: 0.3691 - degree_2_loss: 0.7893 - quality_loss: 0.7726 - inversion_loss: 0.9018 - root_loss: 0.7115 - key_ac 32/150 [=====>........................] - ETA: 1:00 - loss: 4.0229 - key_loss: 0.4792 - degree_1_loss: 0.3688 - degree_2_loss: 0.7893 - quality_loss: 0.7724 - inversion_loss: 0.9018 - root_loss: 0.7113 - key_ac 33/150 [=====>........................] - ETA: 59s - loss: 4.0224 - key_loss: 0.4789 - degree_1_loss: 0.3686 - degree_2_loss: 0.7895 - quality_loss: 0.7725 - inversion_loss: 0.9018 - root_loss: 0.7112 - key_acc 34/150 [=====>........................] - ETA: 59s - loss: 4.0220 - key_loss: 0.4787 - degree_1_loss: 0.3683 - degree_2_loss: 0.7896 - quality_loss: 0.7725 - inversion_loss: 0.9020 - root_loss: 0.7110 - key_acc 35/150 [======>.......................] - ETA: 58s - loss: 4.0212 - key_loss: 0.4784 - degree_1_loss: 0.3680 - degree_2_loss: 0.7896 - quality_loss: 0.7724 - inversion_loss: 0.9020 - root_loss: 0.7107 - key_acc 36/150 [======>.......................] - ETA: 58s - loss: 4.0204 - key_loss: 0.4782 - degree_1_loss: 0.3678 - degree_2_loss: 0.7896 - quality_loss: 0.7723 - inversion_loss: 0.9021 - root_loss: 0.7104 - key_acc 37/150 [======>.......................] - ETA: 57s - loss: 4.0195 - key_loss: 0.4779 - degree_1_loss: 0.3675 - degree_2_loss: 0.7896 - quality_loss: 0.7723 - inversion_loss: 0.9021 - root_loss: 0.7100 - key_acc 38/150 [======>.......................] - ETA: 57s - loss: 4.0185 - key_loss: 0.4777 - degree_1_loss: 0.3673 - degree_2_loss: 0.7896 - quality_loss: 0.7722 - inversion_loss: 0.9021 - root_loss: 0.7096 - key_acc 39/150 [======>.......................] - ETA: 56s - loss: 4.0172 - key_loss: 0.4773 - degree_1_loss: 0.3669 - degree_2_loss: 0.7896 - quality_loss: 0.7721 - inversion_loss: 0.9021 - root_loss: 0.7092 - key_acc 40/150 [=======>......................] - ETA: 56s - loss: 4.0161 - key_loss: 0.4770 - degree_1_loss: 0.3666 - degree_2_loss: 0.7896 - quality_loss: 0.7719 - inversion_loss: 0.9022 - root_loss: 0.7088 - key_acc 41/150 [=======>......................] - ETA: 55s - loss: 4.0151 - key_loss: 0.4768 - degree_1_loss: 0.3664 - degree_2_loss: 0.7895 - quality_loss: 0.7718 - inversion_loss: 0.9021 - root_loss: 0.7085 - key_acc 42/150 [=======>......................] - ETA: 55s - loss: 4.0142 - key_loss: 0.4765 - degree_1_loss: 0.3662 - degree_2_loss: 0.7894 - quality_loss: 0.7717 - inversion_loss: 0.9022 - root_loss: 0.7082 - key_acc 43/150 [=======>......................] - ETA: 54s - loss: 4.0134 - key_loss: 0.4762 - degree_1_loss: 0.3660 - degree_2_loss: 0.7893 - quality_loss: 0.7717 - inversion_loss: 0.9023 - root_loss: 0.7080 - key_acc 44/150 [=======>......................] - ETA: 54s - loss: 4.0127 - key_loss: 0.4759 - degree_1_loss: 0.3658 - degree_2_loss: 0.7892 - quality_loss: 0.7716 - inversion_loss: 0.9023 - root_loss: 0.7077 - key_acc 45/150 [========>.....................] - ETA: 53s - loss: 4.0123 - key_loss: 0.4758 - degree_1_loss: 0.3657 - degree_2_loss: 0.7892 - quality_loss: 0.7717 - inversion_loss: 0.9024 - root_loss: 0.7075 - key_acc 46/150 [========>.....................] - ETA: 53s - loss: 4.0118 - key_loss: 0.4756 - degree_1_loss: 0.3655 - degree_2_loss: 0.7892 - quality_loss: 0.7717 - inversion_loss: 0.9025 - root_loss: 0.7073 - key_acc 47/150 [========>.....................] - ETA: 52s - loss: 4.0113 - key_loss: 0.4754 - degree_1_loss: 0.3653 - degree_2_loss: 0.7892 - quality_loss: 0.7717 - inversion_loss: 0.9026 - root_loss: 0.7071 - key_acc 48/150 [========>.....................] - ETA: 51s - loss: 4.0106 - key_loss: 0.4751 - degree_1_loss: 0.3651 - degree_2_loss: 0.7891 - quality_loss: 0.7717 - inversion_loss: 0.9027 - root_loss: 0.7069 - key_acc 49/150 [========>.....................] - ETA: 51s - loss: 4.0100 - key_loss: 0.4749 - degree_1_loss: 0.3650 - degree_2_loss: 0.7890 - quality_loss: 0.7717 - inversion_loss: 0.9028 - root_loss: 0.7067 - key_acc 50/150 [=========>....................] - ETA: 50s - loss: 4.0091 - key_loss: 0.4746 - degree_1_loss: 0.3648 - degree_2_loss: 0.7890 - quality_loss: 0.7716 - inversion_loss: 0.9028 - root_loss: 0.7064 - key_acc 51/150 [=========>....................] - ETA: 50s - loss: 4.0082 - key_loss: 0.4743 - degree_1_loss: 0.3646 - degree_2_loss: 0.7889 - quality_loss: 0.7715 - inversion_loss: 0.9028 - root_loss: 0.7061 - key_acc 52/150 [=========>....................] - ETA: 49s - loss: 4.0075 - key_loss: 0.4740 - degree_1_loss: 0.3644 - degree_2_loss: 0.7889 - quality_loss: 0.7714 - inversion_loss: 0.9029 - root_loss: 0.7059 - key_acc 53/150 [=========>....................] - ETA: 49s - loss: 4.0069 - key_loss: 0.4738 - degree_1_loss: 0.3643 - degree_2_loss: 0.7888 - quality_loss: 0.7714 - inversion_loss: 0.9029 - root_loss: 0.7057 - key_acc 54/150 [=========>....................] - ETA: 48s - loss: 4.0063 - key_loss: 0.4734 - degree_1_loss: 0.3642 - degree_2_loss: 0.7888 - quality_loss: 0.7713 - inversion_loss: 0.9030 - root_loss: 0.7055 - key_acc 55/150 [==========>...................] - ETA: 48s - loss: 4.0059 - key_loss: 0.4732 - degree_1_loss: 0.3641 - degree_2_loss: 0.7888 - quality_loss: 0.7713 - inversion_loss: 0.9031 - root_loss: 0.7054 - key_acc 56/150 [==========>...................] - ETA: 47s - loss: 4.0057 - key_loss: 0.4731 - degree_1_loss: 0.3640 - degree_2_loss: 0.7889 - quality_loss: 0.7713 - inversion_loss: 0.9033 - root_loss: 0.7052 - key_acc 57/150 [==========>...................] - ETA: 47s - loss: 4.0054 - key_loss: 0.4729 - degree_1_loss: 0.3638 - degree_2_loss: 0.7889 - quality_loss: 0.7713 - inversion_loss: 0.9033 - root_loss: 0.7051 - key_acc 58/150 [==========>...................] - ETA: 46s - loss: 4.0052 - key_loss: 0.4728 - degree_1_loss: 0.3637 - degree_2_loss: 0.7890 - quality_loss: 0.7713 - inversion_loss: 0.9034 - root_loss: 0.7050 - key_acc 59/150 [==========>...................] - ETA: 46s - loss: 4.0050 - key_loss: 0.4727 - degree_1_loss: 0.3636 - degree_2_loss: 0.7891 - quality_loss: 0.7713 - inversion_loss: 0.9035 - root_loss: 0.7049 - key_acc 60/150 [===========>..................] - ETA: 45s - loss: 4.0050 - key_loss: 0.4726 - degree_1_loss: 0.3635 - degree_2_loss: 0.7892 - quality_loss: 0.7713 - inversion_loss: 0.9036 - root_loss: 0.7048 - key_acc 61/150 [===========>..................] - ETA: 45s - loss: 4.0050 - key_loss: 0.4725 - degree_1_loss: 0.3634 - degree_2_loss: 0.7893 - quality_loss: 0.7714 - inversion_loss: 0.9037 - root_loss: 0.7047 - key_acc 62/150 [===========>..................] - ETA: 44s - loss: 4.0048 - key_loss: 0.4724 - degree_1_loss: 0.3633 - degree_2_loss: 0.7894 - quality_loss: 0.7714 - inversion_loss: 0.9037 - root_loss: 0.7046 - key_acc 63/150 [===========>..................] - ETA: 44s - loss: 4.0045 - key_loss: 0.4723 - degree_1_loss: 0.3632 - degree_2_loss: 0.7894 - quality_loss: 0.7714 - inversion_loss: 0.9038 - root_loss: 0.7045 - key_acc 64/150 [===========>..................] - ETA: 43s - loss: 4.0042 - key_loss: 0.4721 - degree_1_loss: 0.3630 - degree_2_loss: 0.7895 - quality_loss: 0.7714 - inversion_loss: 0.9038 - root_loss: 0.7044 - key_acc 65/150 [============>.................] - ETA: 43s - loss: 4.0038 - key_loss: 0.4719 - degree_1_loss: 0.3628 - degree_2_loss: 0.7896 - quality_loss: 0.7714 - inversion_loss: 0.9038 - root_loss: 0.7043 - key_acc 66/150 [============>.................] - ETA: 42s - loss: 4.0035 - key_loss: 0.4718 - degree_1_loss: 0.3627 - degree_2_loss: 0.7896 - quality_loss: 0.7714 - inversion_loss: 0.9038 - root_loss: 0.7043 - key_acc 67/150 [============>.................] - ETA: 42s - loss: 4.0031 - key_loss: 0.4716 - degree_1_loss: 0.3625 - degree_2_loss: 0.7897 - quality_loss: 0.7714 - inversion_loss: 0.9038 - root_loss: 0.7042 - key_acc 68/150 [============>.................] - ETA: 41s - loss: 4.0026 - key_loss: 0.4715 - degree_1_loss: 0.3623 - degree_2_loss: 0.7897 - quality_loss: 0.7714 - inversion_loss: 0.9037 - root_loss: 0.7041 - key_acc 69/150 [============>.................] - ETA: 41s - loss: 4.0021 - key_loss: 0.4713 - degree_1_loss: 0.3620 - degree_2_loss: 0.7897 - quality_loss: 0.7713 - inversion_loss: 0.9037 - root_loss: 0.7040 - key_acc 70/150 [=============>................] - ETA: 40s - loss: 4.0016 - key_loss: 0.4711 - degree_1_loss: 0.3618 - degree_2_loss: 0.7897 - quality_loss: 0.7713 - inversion_loss: 0.9037 - root_loss: 0.7039 - key_acc 71/150 [=============>................] - ETA: 40s - loss: 4.0010 - key_loss: 0.4710 - degree_1_loss: 0.3616 - degree_2_loss: 0.7897 - quality_loss: 0.7712 - inversion_loss: 0.9037 - root_loss: 0.7038 - key_acc 72/150 [=============>................] - ETA: 39s - loss: 4.0005 - key_loss: 0.4708 - degree_1_loss: 0.3614 - degree_2_loss: 0.7897 - quality_loss: 0.7712 - inversion_loss: 0.9037 - root_loss: 0.7037 - key_acc 73/150 [=============>................] - ETA: 39s - loss: 4.0001 - key_loss: 0.4707 - degree_1_loss: 0.3612 - degree_2_loss: 0.7897 - quality_loss: 0.7712 - inversion_loss: 0.9037 - root_loss: 0.7036 - key_acc 74/150 [=============>................] - ETA: 38s - loss: 3.9997 - key_loss: 0.4706 - degree_1_loss: 0.3610 - degree_2_loss: 0.7897 - quality_loss: 0.7712 - inversion_loss: 0.9037 - root_loss: 0.7035 - key_acc 75/150 [==============>...............] - ETA: 38s - loss: 3.9993 - key_loss: 0.4705 - degree_1_loss: 0.3608 - degree_2_loss: 0.7898 - quality_loss: 0.7712 - inversion_loss: 0.9037 - root_loss: 0.7034 - key_acc 76/150 [==============>...............] - ETA: 37s - loss: 3.9990 - key_loss: 0.4703 - degree_1_loss: 0.3607 - degree_2_loss: 0.7898 - quality_loss: 0.7711 - inversion_loss: 0.9037 - root_loss: 0.7033 - key_acc 77/150 [==============>...............] - ETA: 37s - loss: 3.9986 - key_loss: 0.4702 - degree_1_loss: 0.3605 - degree_2_loss: 0.7898 - quality_loss: 0.7711 - inversion_loss: 0.9038 - root_loss: 0.7032 - key_acc 78/150 [==============>...............] - ETA: 36s - loss: 3.9982 - key_loss: 0.4701 - degree_1_loss: 0.3604 - degree_2_loss: 0.7897 - quality_loss: 0.7711 - inversion_loss: 0.9038 - root_loss: 0.7031 - key_acc 79/150 [==============>...............] - ETA: 36s - loss: 3.9977 - key_loss: 0.4700 - degree_1_loss: 0.3602 - degree_2_loss: 0.7897 - quality_loss: 0.7710 - inversion_loss: 0.9038 - root_loss: 0.7030 - key_acc 80/150 [===============>..............] - ETA: 35s - loss: 3.9974 - key_loss: 0.4698 - degree_1_loss: 0.3600 - degree_2_loss: 0.7897 - quality_loss: 0.7710 - inversion_loss: 0.9038 - root_loss: 0.7029 - key_acc 81/150 [===============>..............] - ETA: 35s - loss: 3.9970 - key_loss: 0.4696 - degree_1_loss: 0.3598 - degree_2_loss: 0.7897 - quality_loss: 0.7710 - inversion_loss: 0.9038 - root_loss: 0.7029 - key_acc 82/150 [===============>..............] - ETA: 34s - loss: 3.9966 - key_loss: 0.4695 - degree_1_loss: 0.3596 - degree_2_loss: 0.7898 - quality_loss: 0.7710 - inversion_loss: 0.9039 - root_loss: 0.7028 - key_acc 83/150 [===============>..............] - ETA: 34s - loss: 3.9962 - key_loss: 0.4693 - degree_1_loss: 0.3595 - degree_2_loss: 0.7898 - quality_loss: 0.7710 - inversion_loss: 0.9039 - root_loss: 0.7028 - key_acc 84/150 [===============>..............] - ETA: 33s - loss: 3.9957 - key_loss: 0.4691 - degree_1_loss: 0.3593 - degree_2_loss: 0.7898 - quality_loss: 0.7710 - inversion_loss: 0.9039 - root_loss: 0.7027 - key_acc 85/150 [================>.............] - ETA: 33s - loss: 3.9953 - key_loss: 0.4690 - degree_1_loss: 0.3591 - degree_2_loss: 0.7898 - quality_loss: 0.7709 - inversion_loss: 0.9039 - root_loss: 0.7026 - key_acc 86/150 [================>.............] - ETA: 32s - loss: 3.9949 - key_loss: 0.4689 - degree_1_loss: 0.3589 - degree_2_loss: 0.7898 - quality_loss: 0.7709 - inversion_loss: 0.9039 - root_loss: 0.7025 - key_acc 87/150 [================>.............] - ETA: 32s - loss: 3.9946 - key_loss: 0.4687 - degree_1_loss: 0.3588 - degree_2_loss: 0.7898 - quality_loss: 0.7709 - inversion_loss: 0.9039 - root_loss: 0.7024 - key_acc 88/150 [================>.............] - ETA: 31s - loss: 3.9942 - key_loss: 0.4686 - degree_1_loss: 0.3586 - degree_2_loss: 0.7898 - quality_loss: 0.7709 - inversion_loss: 0.9040 - root_loss: 0.7024 - key_acc 89/150 [================>.............] - ETA: 31s - loss: 3.9940 - key_loss: 0.4685 - degree_1_loss: 0.3584 - degree_2_loss: 0.7899 - quality_loss: 0.7709 - inversion_loss: 0.9040 - root_loss: 0.7023 - key_acc 90/150 [=================>............] - ETA: 30s - loss: 3.9937 - key_loss: 0.4684 - degree_1_loss: 0.3583 - degree_2_loss: 0.7899 - quality_loss: 0.7709 - inversion_loss: 0.9040 - root_loss: 0.7022 - key_acc 91/150 [=================>............] - ETA: 30s - loss: 3.9934 - key_loss: 0.4683 - degree_1_loss: 0.3581 - degree_2_loss: 0.7899 - quality_loss: 0.7708 - inversion_loss: 0.9040 - root_loss: 0.7022 - key_acc 92/150 [=================>............] - ETA: 29s - loss: 3.9931 - key_loss: 0.4682 - degree_1_loss: 0.3580 - degree_2_loss: 0.7899 - quality_loss: 0.7709 - inversion_loss: 0.9040 - root_loss: 0.7021 - key_acc 93/150 [=================>............] - ETA: 29s - loss: 3.9928 - key_loss: 0.4681 - degree_1_loss: 0.3579 - degree_2_loss: 0.7900 - quality_loss: 0.7708 - inversion_loss: 0.9041 - root_loss: 0.7020 - key_acc 94/150 [=================>............] - ETA: 28s - loss: 3.9926 - key_loss: 0.4679 - degree_1_loss: 0.3577 - degree_2_loss: 0.7900 - quality_loss: 0.7709 - inversion_loss: 0.9041 - root_loss: 0.7020 - key_acc 95/150 [==================>...........] - ETA: 27s - loss: 3.9923 - key_loss: 0.4678 - degree_1_loss: 0.3576 - degree_2_loss: 0.7900 - quality_loss: 0.7709 - inversion_loss: 0.9041 - root_loss: 0.7019 - key_acc 96/150 [==================>...........] - ETA: 27s - loss: 3.9921 - key_loss: 0.4677 - degree_1_loss: 0.3575 - degree_2_loss: 0.7901 - quality_loss: 0.7709 - inversion_loss: 0.9041 - root_loss: 0.7018 - key_acc 97/150 [==================>...........] - ETA: 26s - loss: 3.9920 - key_loss: 0.4676 - degree_1_loss: 0.3574 - degree_2_loss: 0.7901 - quality_loss: 0.7709 - inversion_loss: 0.9042 - root_loss: 0.7018 - key_acc 98/150 [==================>...........] - ETA: 26s - loss: 3.9919 - key_loss: 0.4675 - degree_1_loss: 0.3573 - degree_2_loss: 0.7902 - quality_loss: 0.7709 - inversion_loss: 0.9042 - root_loss: 0.7017 - key_acc 99/150 [==================>...........] - ETA: 25s - loss: 3.9917 - key_loss: 0.4674 - degree_1_loss: 0.3572 - degree_2_loss: 0.7902 - quality_loss: 0.7710 - inversion_loss: 0.9042 - root_loss: 0.7017 - key_acc100/150 [===================>..........] - ETA: 25s - loss: 3.9914 - key_loss: 0.4673 - degree_1_loss: 0.3571 - degree_2_loss: 0.7902 - quality_loss: 0.7710 - inversion_loss: 0.9042 - root_loss: 0.7016 - key_acc101/150 [===================>..........] - ETA: 24s - loss: 3.9913 - key_loss: 0.4671 - degree_1_loss: 0.3570 - degree_2_loss: 0.7903 - quality_loss: 0.7710 - inversion_loss: 0.9042 - root_loss: 0.7016 - key_acc102/150 [===================>..........] - ETA: 24s - loss: 3.9911 - key_loss: 0.4670 - degree_1_loss: 0.3569 - degree_2_loss: 0.7903 - quality_loss: 0.7710 - inversion_loss: 0.9042 - root_loss: 0.7015 - key_acc103/150 [===================>..........] - ETA: 23s - loss: 3.9909 - key_loss: 0.4670 - degree_1_loss: 0.3568 - degree_2_loss: 0.7904 - quality_loss: 0.7711 - inversion_loss: 0.9042 - root_loss: 0.7015 - key_acc104/150 [===================>..........] - ETA: 23s - loss: 3.9908 - key_loss: 0.4669 - degree_1_loss: 0.3567 - degree_2_loss: 0.7904 - quality_loss: 0.7711 - inversion_loss: 0.9042 - root_loss: 0.7015 - key_acc105/150 [====================>.........] - ETA: 22s - loss: 3.9908 - key_loss: 0.4668 - degree_1_loss: 0.3567 - degree_2_loss: 0.7904 - quality_loss: 0.7712 - inversion_loss: 0.9043 - root_loss: 0.7015 - key_acc106/150 [====================>.........] - ETA: 22s - loss: 3.9908 - key_loss: 0.4668 - degree_1_loss: 0.3566 - degree_2_loss: 0.7905 - quality_loss: 0.7712 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc107/150 [====================>.........] - ETA: 21s - loss: 3.9907 - key_loss: 0.4667 - degree_1_loss: 0.3565 - degree_2_loss: 0.7905 - quality_loss: 0.7712 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc108/150 [====================>.........] - ETA: 21s - loss: 3.9907 - key_loss: 0.4666 - degree_1_loss: 0.3564 - degree_2_loss: 0.7906 - quality_loss: 0.7713 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc109/150 [====================>.........] - ETA: 20s - loss: 3.9906 - key_loss: 0.4666 - degree_1_loss: 0.3564 - degree_2_loss: 0.7907 - quality_loss: 0.7713 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc110/150 [=====================>........] - ETA: 20s - loss: 3.9906 - key_loss: 0.4665 - degree_1_loss: 0.3563 - degree_2_loss: 0.7907 - quality_loss: 0.7714 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc111/150 [=====================>........] - ETA: 19s - loss: 3.9907 - key_loss: 0.4665 - degree_1_loss: 0.3563 - degree_2_loss: 0.7908 - quality_loss: 0.7714 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc112/150 [=====================>........] - ETA: 19s - loss: 3.9907 - key_loss: 0.4665 - degree_1_loss: 0.3562 - degree_2_loss: 0.7909 - quality_loss: 0.7714 - inversion_loss: 0.9043 - root_loss: 0.7014 - key_acc113/150 [=====================>........] - ETA: 18s - loss: 3.9908 - key_loss: 0.4664 - degree_1_loss: 0.3562 - degree_2_loss: 0.7909 - quality_loss: 0.7715 - inversion_loss: 0.9044 - root_loss: 0.7014 - key_acc114/150 [=====================>........] - ETA: 18s - loss: 3.9908 - key_loss: 0.4664 - degree_1_loss: 0.3561 - degree_2_loss: 0.7910 - quality_loss: 0.7715 - inversion_loss: 0.9044 - root_loss: 0.7013 - key_acc115/150 [======================>.......] - ETA: 17s - loss: 3.9908 - key_loss: 0.4664 - degree_1_loss: 0.3561 - degree_2_loss: 0.7911 - quality_loss: 0.7715 - inversion_loss: 0.9044 - root_loss: 0.7013 - key_acc116/150 [======================>.......] - ETA: 17s - loss: 3.9909 - key_loss: 0.4664 - degree_1_loss: 0.3560 - degree_2_loss: 0.7911 - quality_loss: 0.7715 - inversion_loss: 0.9044 - root_loss: 0.7013 - key_acc117/150 [======================>.......] - ETA: 16s - loss: 3.9909 - key_loss: 0.4664 - degree_1_loss: 0.3560 - degree_2_loss: 0.7912 - quality_loss: 0.7716 - inversion_loss: 0.9044 - root_loss: 0.7013 - key_acc118/150 [======================>.......] - ETA: 16s - loss: 3.9909 - key_loss: 0.4664 - degree_1_loss: 0.3560 - degree_2_loss: 0.7912 - quality_loss: 0.7716 - inversion_loss: 0.9044 - root_loss: 0.7013 - key_acc119/150 [======================>.......] - ETA: 15s - loss: 3.9909 - key_loss: 0.4665 - degree_1_loss: 0.3559 - degree_2_loss: 0.7913 - quality_loss: 0.7716 - inversion_loss: 0.9044 - root_loss: 0.7012 - key_acc120/150 [=======================>......] - ETA: 15s - loss: 3.9909 - key_loss: 0.4665 - degree_1_loss: 0.3559 - degree_2_loss: 0.7913 - quality_loss: 0.7716 - inversion_loss: 0.9044 - root_loss: 0.7012 - key_acc121/150 [=======================>......] - ETA: 14s - loss: 3.9910 - key_loss: 0.4665 - degree_1_loss: 0.3558 - degree_2_loss: 0.7914 - quality_loss: 0.7716 - inversion_loss: 0.9044 - root_loss: 0.7012 - key_acc122/150 [=======================>......] - ETA: 14s - loss: 3.9910 - key_loss: 0.4665 - degree_1_loss: 0.3558 - degree_2_loss: 0.7914 - quality_loss: 0.7716 - inversion_loss: 0.9044 - root_loss: 0.7012 - key_acc123/150 [=======================>......] - ETA: 13s - loss: 3.9911 - key_loss: 0.4665 - degree_1_loss: 0.3558 - degree_2_loss: 0.7915 - quality_loss: 0.7717 - inversion_loss: 0.9044 - root_loss: 0.7012 - key_acc124/150 [=======================>......] - ETA: 13s - loss: 3.9911 - key_loss: 0.4665 - degree_1_loss: 0.3557 - degree_2_loss: 0.7916 - quality_loss: 0.7717 - inversion_loss: 0.9045 - root_loss: 0.7012 - key_acc125/150 [========================>.....] - ETA: 12s - loss: 3.9911 - key_loss: 0.4665 - degree_1_loss: 0.3557 - degree_2_loss: 0.7916 - quality_loss: 0.7717 - inversion_loss: 0.9045 - root_loss: 0.7012 - key_acc126/150 [========================>.....] - ETA: 12s - loss: 3.9912 - key_loss: 0.4665 - degree_1_loss: 0.3556 - degree_2_loss: 0.7917 - quality_loss: 0.7717 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_acc127/150 [========================>.....] - ETA: 11s - loss: 3.9912 - key_loss: 0.4665 - degree_1_loss: 0.3556 - degree_2_loss: 0.7917 - quality_loss: 0.7718 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_acc128/150 [========================>.....] - ETA: 11s - loss: 3.9912 - key_loss: 0.4665 - degree_1_loss: 0.3556 - degree_2_loss: 0.7917 - quality_loss: 0.7718 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_acc129/150 [========================>.....] - ETA: 10s - loss: 3.9912 - key_loss: 0.4665 - degree_1_loss: 0.3555 - degree_2_loss: 0.7918 - quality_loss: 0.7718 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_acc130/150 [=========================>....] - ETA: 10s - loss: 3.9913 - key_loss: 0.4666 - degree_1_loss: 0.3555 - degree_2_loss: 0.7918 - quality_loss: 0.7718 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_acc131/150 [=========================>....] - ETA: 9s - loss: 3.9913 - key_loss: 0.4666 - degree_1_loss: 0.3555 - degree_2_loss: 0.7919 - quality_loss: 0.7718 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_accu132/150 [=========================>....] - ETA: 9s - loss: 3.9914 - key_loss: 0.4666 - degree_1_loss: 0.3554 - degree_2_loss: 0.7920 - quality_loss: 0.7719 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_accu133/150 [=========================>....] - ETA: 8s - loss: 3.9915 - key_loss: 0.4666 - degree_1_loss: 0.3554 - degree_2_loss: 0.7920 - quality_loss: 0.7719 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_accu134/150 [=========================>....] - ETA: 8s - loss: 3.9915 - key_loss: 0.4666 - degree_1_loss: 0.3554 - degree_2_loss: 0.7921 - quality_loss: 0.7719 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_accu135/150 [==========================>...] - ETA: 7s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3553 - degree_2_loss: 0.7921 - quality_loss: 0.7720 - inversion_loss: 0.9044 - root_loss: 0.7011 - key_accu136/150 [==========================>...] - ETA: 7s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3553 - degree_2_loss: 0.7922 - quality_loss: 0.7720 - inversion_loss: 0.9043 - root_loss: 0.7010 - key_accu137/150 [==========================>...] - ETA: 6s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3553 - degree_2_loss: 0.7922 - quality_loss: 0.7720 - inversion_loss: 0.9043 - root_loss: 0.7010 - key_accu138/150 [==========================>...] - ETA: 6s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3552 - degree_2_loss: 0.7923 - quality_loss: 0.7720 - inversion_loss: 0.9043 - root_loss: 0.7010 - key_accu139/150 [==========================>...] - ETA: 5s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3552 - degree_2_loss: 0.7923 - quality_loss: 0.7720 - inversion_loss: 0.9042 - root_loss: 0.7010 - key_accu140/150 [===========================>..] - ETA: 5s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3552 - degree_2_loss: 0.7924 - quality_loss: 0.7720 - inversion_loss: 0.9042 - root_loss: 0.7010 - key_accu141/150 [===========================>..] - ETA: 4s - loss: 3.9914 - key_loss: 0.4667 - degree_1_loss: 0.3551 - degree_2_loss: 0.7924 - quality_loss: 0.7720 - inversion_loss: 0.9042 - root_loss: 0.7010 - key_accu142/150 [===========================>..] - ETA: 4s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3551 - degree_2_loss: 0.7925 - quality_loss: 0.7721 - inversion_loss: 0.9042 - root_loss: 0.7009 - key_accu143/150 [===========================>..] - ETA: 3s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3551 - degree_2_loss: 0.7925 - quality_loss: 0.7721 - inversion_loss: 0.9041 - root_loss: 0.7009 - key_accu144/150 [===========================>..] - ETA: 3s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3551 - degree_2_loss: 0.7926 - quality_loss: 0.7721 - inversion_loss: 0.9041 - root_loss: 0.7009 - key_accu145/150 [============================>.] - ETA: 2s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3550 - degree_2_loss: 0.7926 - quality_loss: 0.7721 - inversion_loss: 0.9041 - root_loss: 0.7009 - key_accu146/150 [============================>.] - ETA: 2s - loss: 3.9915 - key_loss: 0.4667 - degree_1_loss: 0.3550 - degree_2_loss: 0.7926 - quality_loss: 0.7721 - inversion_loss: 0.9041 - root_loss: 0.7009 - key_accu147/150 [============================>.] - ETA: 1s - loss: 3.9914 - key_loss: 0.4667 - degree_1_loss: 0.3549 - degree_2_loss: 0.7927 - quality_loss: 0.7721 - inversion_loss: 0.9040 - root_loss: 0.7009 - key_accu148/150 [============================>.] - ETA: 1s - loss: 3.9913 - key_loss: 0.4667 - degree_1_loss: 0.3549 - degree_2_loss: 0.7927 - quality_loss: 0.7721 - inversion_loss: 0.9040 - root_loss: 0.7009 - key_accu149/150 [============================>.] - ETA: 0s - loss: 3.9912 - key_loss: 0.4667 - degree_1_loss: 0.3548 - degree_2_loss: 0.7927 - quality_loss: 0.7721 - inversion_loss: 0.9040 - root_loss: 0.7009 - key_accu150/150 [==============================] - ETA: 0s - loss: 3.9912 - key_loss: 0.4667 - degree_1_loss: 0.3548 - degree_2_loss: 0.7928 - quality_loss: 0.7721 - inversion_loss: 0.9039 - root_loss: 0.7009 - key_accu150/150 [==============================] - 77s 512ms/step - loss: 3.9911 - key_loss: 0.4667 - degree_1_loss: 0.3547 - degree_2_loss: 0.7928 - quality_loss: 0.7721 - inversion_loss: 0.9039 - root_loss: 0.7009 - key_accuracy: 0.8440 - degree_1_accuracy: 0.9059 - degree_2_accuracy: 0.7043 - quality_accuracy: 0.7164 - inversion_accuracy: 0.6022 - root_accuracy: 0.7589 - val_loss: 4.3714 - val_key_loss: 0.6944 - val_degree_1_loss: 0.3546 - val_degree_2_loss: 0.8736 - val_quality_loss: 0.8298 - val_inversion_loss: 0.9094 - val_root_loss: 0.7096 - val_key_accuracy: 0.7618 - val_degree_1_accuracy: 0.9119 - val_degree_2_accuracy: 0.6805 - val_quality_accuracy: 0.7015 - val_inversion_accuracy: 0.6232 - val_root_accuracy: 0.7683

loss: 3.9911
key_loss: 0.4667
degree_1_loss: 0.3547
degree_2_loss: 0.7928
quality_loss: 0.7721
inversion_loss: 0.9039
root_loss: 0.7009

key_accuracy: 0.8440
degree_1_accuracy: 0.9059
degree_2_accuracy: 0.7043
quality_accuracy: 0.7164
inversion_accuracy: 0.6022
root_accuracy: 0.7589

val_loss: 4.3714
val_key_loss: 0.6944
val_degree_1_loss: 0.3546
val_degree_2_loss: 0.8736
val_quality_loss: 0.8298
val_inversion_loss: 0.9094
val_root_loss: 0.7096

val_key_accuracy: 0.7618
val_degree_1_accuracy: 0.9119
val_degree_2_accuracy: 0.6805
val_quality_accuracy: 0.7015
val_inversion_accuracy: 0.6232
val_root_accuracy: 0.7683

python train.py --model gru --input pitch_hybrid_cut

Running with N_OCTAVES==1 (only middle C; 19 features).

Epoch 26/100
  1/124 [..............................] - ETA: 1:10 - loss: 4.0788 - key_loss: 0.4024 - degree_1_loss: 0.4734 - degree_2_loss: 0.7715 - quality_loss: 0.8577 - inversion_loss: 0.9355 - root_loss: 0.6382 - key_ac  2/124 [..............................] - ETA: 1:03 - loss: 3.9856 - key_loss: 0.3825 - degree_1_loss: 0.4339 - degree_2_loss: 0.7648 - quality_loss: 0.8270 - inversion_loss: 0.9169 - root_loss: 0.6605 - key_ac  3/124 [..............................] - ETA: 1:01 - loss: 3.9892 - key_loss: 0.3874 - degree_1_loss: 0.4222 - degree_2_loss: 0.7734 - quality_loss: 0.8143 - inversion_loss: 0.9144 - root_loss: 0.6775 - key_ac  4/124 [..............................] - ETA: 1:01 - loss: 4.0250 - key_loss: 0.3998 - degree_1_loss: 0.4121 - degree_2_loss: 0.7842 - quality_loss: 0.8164 - inversion_loss: 0.9193 - root_loss: 0.6931 - key_ac  5/124 [>.............................] - ETA: 1:00 - loss: 4.0426 - key_loss: 0.4074 - degree_1_loss: 0.4030 - degree_2_loss: 0.7919 - quality_loss: 0.8184 - inversion_loss: 0.9194 - root_loss: 0.7025 - key_ac  6/124 [>.............................] - ETA: 1:00 - loss: 4.0615 - key_loss: 0.4134 - degree_1_loss: 0.3949 - degree_2_loss: 0.7991 - quality_loss: 0.8220 - inversion_loss: 0.9195 - root_loss: 0.7124 - key_ac  7/124 [>.............................] - ETA: 1:00 - loss: 4.0749 - key_loss: 0.4156 - degree_1_loss: 0.3890 - degree_2_loss: 0.8050 - quality_loss: 0.8245 - inversion_loss: 0.9228 - root_loss: 0.7179 - key_ac  8/124 [>.............................] - ETA: 59s - loss: 4.0881 - key_loss: 0.4200 - degree_1_loss: 0.3866 - degree_2_loss: 0.8091 - quality_loss: 0.8256 - inversion_loss: 0.9271 - root_loss: 0.7196 - key_acc  9/124 [=>............................] - ETA: 59s - loss: 4.1038 - key_loss: 0.4231 - degree_1_loss: 0.3851 - degree_2_loss: 0.8136 - quality_loss: 0.8291 - inversion_loss: 0.9309 - root_loss: 0.7221 - key_acc 10/124 [=>............................] - ETA: 58s - loss: 4.1212 - key_loss: 0.4265 - degree_1_loss: 0.3839 - degree_2_loss: 0.8183 - quality_loss: 0.8329 - inversion_loss: 0.9340 - root_loss: 0.7255 - key_acc 11/124 [=>............................] - ETA: 57s - loss: 4.1335 - key_loss: 0.4289 - degree_1_loss: 0.3826 - degree_2_loss: 0.8219 - quality_loss: 0.8356 - inversion_loss: 0.9368 - root_loss: 0.7276 - key_acc 12/124 [=>............................] - ETA: 57s - loss: 4.1432 - key_loss: 0.4314 - degree_1_loss: 0.3813 - degree_2_loss: 0.8250 - quality_loss: 0.8380 - inversion_loss: 0.9384 - root_loss: 0.7291 - key_acc 13/124 [==>...........................] - ETA: 56s - loss: 4.1527 - key_loss: 0.4340 - degree_1_loss: 0.3806 - degree_2_loss: 0.8284 - quality_loss: 0.8396 - inversion_loss: 0.9396 - root_loss: 0.7305 - key_acc 14/124 [==>...........................] - ETA: 56s - loss: 4.1602 - key_loss: 0.4362 - degree_1_loss: 0.3799 - degree_2_loss: 0.8314 - quality_loss: 0.8406 - inversion_loss: 0.9406 - root_loss: 0.7314 - key_acc 15/124 [==>...........................] - ETA: 55s - loss: 4.1673 - key_loss: 0.4381 - degree_1_loss: 0.3795 - degree_2_loss: 0.8341 - quality_loss: 0.8412 - inversion_loss: 0.9419 - root_loss: 0.7325 - key_acc 16/124 [==>...........................] - ETA: 55s - loss: 4.1740 - key_loss: 0.4398 - degree_1_loss: 0.3795 - degree_2_loss: 0.8363 - quality_loss: 0.8417 - inversion_loss: 0.9431 - root_loss: 0.7335 - key_acc 17/124 [===>..........................] - ETA: 54s - loss: 4.1806 - key_loss: 0.4420 - degree_1_loss: 0.3798 - degree_2_loss: 0.8382 - quality_loss: 0.8420 - inversion_loss: 0.9443 - root_loss: 0.7344 - key_acc 18/124 [===>..........................] - ETA: 54s - loss: 4.1860 - key_loss: 0.4445 - degree_1_loss: 0.3799 - degree_2_loss: 0.8399 - quality_loss: 0.8421 - inversion_loss: 0.9447 - root_loss: 0.7350 - key_acc 19/124 [===>..........................] - ETA: 53s - loss: 4.1904 - key_loss: 0.4466 - degree_1_loss: 0.3800 - degree_2_loss: 0.8413 - quality_loss: 0.8419 - inversion_loss: 0.9453 - root_loss: 0.7353 - key_acc 20/124 [===>..........................] - ETA: 53s - loss: 4.1929 - key_loss: 0.4481 - degree_1_loss: 0.3802 - degree_2_loss: 0.8423 - quality_loss: 0.8416 - inversion_loss: 0.9455 - root_loss: 0.7353 - key_acc 21/124 [====>.........................] - ETA: 52s - loss: 4.1944 - key_loss: 0.4494 - degree_1_loss: 0.3801 - degree_2_loss: 0.8430 - quality_loss: 0.8411 - inversion_loss: 0.9457 - root_loss: 0.7350 - key_acc 22/124 [====>.........................] - ETA: 52s - loss: 4.1945 - key_loss: 0.4506 - degree_1_loss: 0.3800 - degree_2_loss: 0.8435 - quality_loss: 0.8404 - inversion_loss: 0.9456 - root_loss: 0.7345 - key_acc 23/124 [====>.........................] - ETA: 51s - loss: 4.1932 - key_loss: 0.4515 - degree_1_loss: 0.3797 - degree_2_loss: 0.8436 - quality_loss: 0.8396 - inversion_loss: 0.9451 - root_loss: 0.7338 - key_acc 24/124 [====>.........................] - ETA: 51s - loss: 4.1922 - key_loss: 0.4522 - degree_1_loss: 0.3794 - degree_2_loss: 0.8439 - quality_loss: 0.8389 - inversion_loss: 0.9447 - root_loss: 0.7332 - key_acc 25/124 [=====>........................] - ETA: 50s - loss: 4.1915 - key_loss: 0.4528 - degree_1_loss: 0.3790 - degree_2_loss: 0.8440 - quality_loss: 0.8385 - inversion_loss: 0.9443 - root_loss: 0.7328 - key_acc 26/124 [=====>........................] - ETA: 50s - loss: 4.1910 - key_loss: 0.4534 - degree_1_loss: 0.3788 - degree_2_loss: 0.8442 - quality_loss: 0.8380 - inversion_loss: 0.9441 - root_loss: 0.7325 - key_acc 27/124 [=====>........................] - ETA: 49s - loss: 4.1901 - key_loss: 0.4541 - degree_1_loss: 0.3785 - degree_2_loss: 0.8441 - quality_loss: 0.8374 - inversion_loss: 0.9438 - root_loss: 0.7321 - key_acc 28/124 [=====>........................] - ETA: 48s - loss: 4.1897 - key_loss: 0.4549 - degree_1_loss: 0.3783 - degree_2_loss: 0.8442 - quality_loss: 0.8370 - inversion_loss: 0.9436 - root_loss: 0.7317 - key_acc 29/124 [======>.......................] - ETA: 48s - loss: 4.1892 - key_loss: 0.4556 - degree_1_loss: 0.3781 - degree_2_loss: 0.8441 - quality_loss: 0.8365 - inversion_loss: 0.9434 - root_loss: 0.7314 - key_acc 30/124 [======>.......................] - ETA: 47s - loss: 4.1888 - key_loss: 0.4561 - degree_1_loss: 0.3779 - degree_2_loss: 0.8441 - quality_loss: 0.8362 - inversion_loss: 0.9433 - root_loss: 0.7312 - key_acc 31/124 [======>.......................] - ETA: 47s - loss: 4.1888 - key_loss: 0.4568 - degree_1_loss: 0.3777 - degree_2_loss: 0.8442 - quality_loss: 0.8360 - inversion_loss: 0.9432 - root_loss: 0.7310 - key_acc 32/124 [======>.......................] - ETA: 46s - loss: 4.1897 - key_loss: 0.4574 - degree_1_loss: 0.3776 - degree_2_loss: 0.8443 - quality_loss: 0.8359 - inversion_loss: 0.9433 - root_loss: 0.7311 - key_acc 33/124 [======>.......................] - ETA: 46s - loss: 4.1905 - key_loss: 0.4581 - degree_1_loss: 0.3774 - degree_2_loss: 0.8445 - quality_loss: 0.8359 - inversion_loss: 0.9434 - root_loss: 0.7311 - key_acc 34/124 [=======>......................] - ETA: 45s - loss: 4.1910 - key_loss: 0.4589 - degree_1_loss: 0.3771 - degree_2_loss: 0.8446 - quality_loss: 0.8358 - inversion_loss: 0.9434 - root_loss: 0.7311 - key_acc 35/124 [=======>......................] - ETA: 45s - loss: 4.1912 - key_loss: 0.4595 - degree_1_loss: 0.3768 - degree_2_loss: 0.8447 - quality_loss: 0.8357 - inversion_loss: 0.9434 - root_loss: 0.7310 - key_acc 36/124 [=======>......................] - ETA: 44s - loss: 4.1913 - key_loss: 0.4602 - degree_1_loss: 0.3764 - degree_2_loss: 0.8447 - quality_loss: 0.8355 - inversion_loss: 0.9435 - root_loss: 0.7310 - key_acc 37/124 [=======>......................] - ETA: 44s - loss: 4.1916 - key_loss: 0.4608 - degree_1_loss: 0.3760 - degree_2_loss: 0.8448 - quality_loss: 0.8354 - inversion_loss: 0.9436 - root_loss: 0.7310 - key_acc 38/124 [========>.....................] - ETA: 43s - loss: 4.1918 - key_loss: 0.4614 - degree_1_loss: 0.3757 - degree_2_loss: 0.8448 - quality_loss: 0.8352 - inversion_loss: 0.9437 - root_loss: 0.7310 - key_acc 39/124 [========>.....................] - ETA: 43s - loss: 4.1922 - key_loss: 0.4619 - degree_1_loss: 0.3754 - degree_2_loss: 0.8449 - quality_loss: 0.8351 - inversion_loss: 0.9440 - root_loss: 0.7310 - key_acc 40/124 [========>.....................] - ETA: 42s - loss: 4.1924 - key_loss: 0.4623 - degree_1_loss: 0.3751 - degree_2_loss: 0.8449 - quality_loss: 0.8350 - inversion_loss: 0.9441 - root_loss: 0.7310 - key_acc 41/124 [========>.....................] - ETA: 42s - loss: 4.1927 - key_loss: 0.4627 - degree_1_loss: 0.3748 - degree_2_loss: 0.8450 - quality_loss: 0.8349 - inversion_loss: 0.9443 - root_loss: 0.7310 - key_acc 42/124 [=========>....................] - ETA: 41s - loss: 4.1931 - key_loss: 0.4631 - degree_1_loss: 0.3745 - degree_2_loss: 0.8450 - quality_loss: 0.8349 - inversion_loss: 0.9444 - root_loss: 0.7311 - key_acc 43/124 [=========>....................] - ETA: 41s - loss: 4.1935 - key_loss: 0.4635 - degree_1_loss: 0.3743 - degree_2_loss: 0.8451 - quality_loss: 0.8349 - inversion_loss: 0.9446 - root_loss: 0.7312 - key_acc 44/124 [=========>....................] - ETA: 40s - loss: 4.1940 - key_loss: 0.4638 - degree_1_loss: 0.3741 - degree_2_loss: 0.8452 - quality_loss: 0.8348 - inversion_loss: 0.9447 - root_loss: 0.7313 - key_acc 45/124 [=========>....................] - ETA: 40s - loss: 4.1943 - key_loss: 0.4641 - degree_1_loss: 0.3739 - degree_2_loss: 0.8453 - quality_loss: 0.8348 - inversion_loss: 0.9448 - root_loss: 0.7314 - key_acc 46/124 [==========>...................] - ETA: 39s - loss: 4.1944 - key_loss: 0.4644 - degree_1_loss: 0.3736 - degree_2_loss: 0.8452 - quality_loss: 0.8347 - inversion_loss: 0.9449 - root_loss: 0.7315 - key_acc 47/124 [==========>...................] - ETA: 39s - loss: 4.1947 - key_loss: 0.4648 - degree_1_loss: 0.3734 - degree_2_loss: 0.8452 - quality_loss: 0.8346 - inversion_loss: 0.9451 - root_loss: 0.7316 - key_acc 48/124 [==========>...................] - ETA: 38s - loss: 4.1948 - key_loss: 0.4650 - degree_1_loss: 0.3732 - degree_2_loss: 0.8452 - quality_loss: 0.8344 - inversion_loss: 0.9452 - root_loss: 0.7317 - key_acc 49/124 [==========>...................] - ETA: 38s - loss: 4.1950 - key_loss: 0.4653 - degree_1_loss: 0.3730 - degree_2_loss: 0.8452 - quality_loss: 0.8343 - inversion_loss: 0.9454 - root_loss: 0.7318 - key_acc 50/124 [===========>..................] - ETA: 37s - loss: 4.1954 - key_loss: 0.4655 - degree_1_loss: 0.3729 - degree_2_loss: 0.8452 - quality_loss: 0.8343 - inversion_loss: 0.9456 - root_loss: 0.7320 - key_acc 51/124 [===========>..................] - ETA: 37s - loss: 4.1957 - key_loss: 0.4657 - degree_1_loss: 0.3728 - degree_2_loss: 0.8452 - quality_loss: 0.8342 - inversion_loss: 0.9457 - root_loss: 0.7322 - key_acc 52/124 [===========>..................] - ETA: 36s - loss: 4.1961 - key_loss: 0.4660 - degree_1_loss: 0.3726 - degree_2_loss: 0.8451 - quality_loss: 0.8342 - inversion_loss: 0.9459 - root_loss: 0.7323 - key_acc 53/124 [===========>..................] - ETA: 36s - loss: 4.1966 - key_loss: 0.4663 - degree_1_loss: 0.3725 - degree_2_loss: 0.8451 - quality_loss: 0.8341 - inversion_loss: 0.9460 - root_loss: 0.7325 - key_acc 54/124 [============>.................] - ETA: 35s - loss: 4.1969 - key_loss: 0.4667 - degree_1_loss: 0.3724 - degree_2_loss: 0.8450 - quality_loss: 0.8340 - inversion_loss: 0.9461 - root_loss: 0.7326 - key_acc 55/124 [============>.................] - ETA: 35s - loss: 4.1971 - key_loss: 0.4670 - degree_1_loss: 0.3723 - degree_2_loss: 0.8449 - quality_loss: 0.8339 - inversion_loss: 0.9462 - root_loss: 0.7327 - key_acc 56/124 [============>.................] - ETA: 34s - loss: 4.1973 - key_loss: 0.4672 - degree_1_loss: 0.3722 - degree_2_loss: 0.8448 - quality_loss: 0.8338 - inversion_loss: 0.9463 - root_loss: 0.7328 - key_acc 57/124 [============>.................] - ETA: 34s - loss: 4.1974 - key_loss: 0.4675 - degree_1_loss: 0.3722 - degree_2_loss: 0.8447 - quality_loss: 0.8338 - inversion_loss: 0.9463 - root_loss: 0.7329 - key_acc 58/124 [=============>................] - ETA: 33s - loss: 4.1975 - key_loss: 0.4678 - degree_1_loss: 0.3721 - degree_2_loss: 0.8446 - quality_loss: 0.8336 - inversion_loss: 0.9464 - root_loss: 0.7329 - key_acc 59/124 [=============>................] - ETA: 33s - loss: 4.1975 - key_loss: 0.4681 - degree_1_loss: 0.3721 - degree_2_loss: 0.8445 - quality_loss: 0.8335 - inversion_loss: 0.9464 - root_loss: 0.7329 - key_acc 60/124 [=============>................] - ETA: 32s - loss: 4.1975 - key_loss: 0.4684 - degree_1_loss: 0.3720 - degree_2_loss: 0.8443 - quality_loss: 0.8334 - inversion_loss: 0.9465 - root_loss: 0.7329 - key_acc 61/124 [=============>................] - ETA: 32s - loss: 4.1974 - key_loss: 0.4686 - degree_1_loss: 0.3719 - degree_2_loss: 0.8442 - quality_loss: 0.8333 - inversion_loss: 0.9466 - root_loss: 0.7328 - key_acc 62/124 [==============>...............] - ETA: 31s - loss: 4.1974 - key_loss: 0.4689 - degree_1_loss: 0.3719 - degree_2_loss: 0.8441 - quality_loss: 0.8331 - inversion_loss: 0.9466 - root_loss: 0.7328 - key_acc 63/124 [==============>...............] - ETA: 31s - loss: 4.1972 - key_loss: 0.4691 - degree_1_loss: 0.3718 - degree_2_loss: 0.8440 - quality_loss: 0.8330 - inversion_loss: 0.9467 - root_loss: 0.7327 - key_acc 64/124 [==============>...............] - ETA: 30s - loss: 4.1971 - key_loss: 0.4693 - degree_1_loss: 0.3717 - degree_2_loss: 0.8438 - quality_loss: 0.8329 - inversion_loss: 0.9467 - root_loss: 0.7327 - key_acc 65/124 [==============>...............] - ETA: 30s - loss: 4.1970 - key_loss: 0.4696 - degree_1_loss: 0.3716 - degree_2_loss: 0.8437 - quality_loss: 0.8328 - inversion_loss: 0.9468 - root_loss: 0.7326 - key_acc 66/124 [==============>...............] - ETA: 29s - loss: 4.1969 - key_loss: 0.4698 - degree_1_loss: 0.3715 - degree_2_loss: 0.8435 - quality_loss: 0.8326 - inversion_loss: 0.9468 - root_loss: 0.7326 - key_acc 67/124 [===============>..............] - ETA: 29s - loss: 4.1967 - key_loss: 0.4700 - degree_1_loss: 0.3715 - degree_2_loss: 0.8433 - quality_loss: 0.8325 - inversion_loss: 0.9469 - root_loss: 0.7325 - key_acc 68/124 [===============>..............] - ETA: 28s - loss: 4.1964 - key_loss: 0.4702 - degree_1_loss: 0.3714 - degree_2_loss: 0.8432 - quality_loss: 0.8323 - inversion_loss: 0.9470 - root_loss: 0.7324 - key_acc 69/124 [===============>..............] - ETA: 28s - loss: 4.1961 - key_loss: 0.4704 - degree_1_loss: 0.3713 - degree_2_loss: 0.8430 - quality_loss: 0.8322 - inversion_loss: 0.9470 - root_loss: 0.7323 - key_acc 70/124 [===============>..............] - ETA: 27s - loss: 4.1956 - key_loss: 0.4705 - degree_1_loss: 0.3712 - degree_2_loss: 0.8428 - quality_loss: 0.8320 - inversion_loss: 0.9470 - root_loss: 0.7322 - key_acc 71/124 [================>.............] - ETA: 27s - loss: 4.1951 - key_loss: 0.4706 - degree_1_loss: 0.3711 - degree_2_loss: 0.8426 - quality_loss: 0.8318 - inversion_loss: 0.9469 - root_loss: 0.7321 - key_acc 72/124 [================>.............] - ETA: 26s - loss: 4.1946 - key_loss: 0.4707 - degree_1_loss: 0.3710 - degree_2_loss: 0.8424 - quality_loss: 0.8316 - inversion_loss: 0.9469 - root_loss: 0.7320 - key_acc 73/124 [================>.............] - ETA: 26s - loss: 4.1941 - key_loss: 0.4708 - degree_1_loss: 0.3710 - degree_2_loss: 0.8422 - quality_loss: 0.8314 - inversion_loss: 0.9469 - root_loss: 0.7319 - key_acc 74/124 [================>.............] - ETA: 25s - loss: 4.1938 - key_loss: 0.4709 - degree_1_loss: 0.3709 - degree_2_loss: 0.8420 - quality_loss: 0.8312 - inversion_loss: 0.9470 - root_loss: 0.7318 - key_acc 75/124 [=================>............] - ETA: 25s - loss: 4.1934 - key_loss: 0.4710 - degree_1_loss: 0.3708 - degree_2_loss: 0.8419 - quality_loss: 0.8310 - inversion_loss: 0.9470 - root_loss: 0.7317 - key_acc 76/124 [=================>............] - ETA: 24s - loss: 4.1930 - key_loss: 0.4711 - degree_1_loss: 0.3707 - degree_2_loss: 0.8417 - quality_loss: 0.8308 - inversion_loss: 0.9470 - root_loss: 0.7317 - key_acc 77/124 [=================>............] - ETA: 24s - loss: 4.1926 - key_loss: 0.4712 - degree_1_loss: 0.3706 - degree_2_loss: 0.8416 - quality_loss: 0.8307 - inversion_loss: 0.9469 - root_loss: 0.7316 - key_acc 78/124 [=================>............] - ETA: 23s - loss: 4.1921 - key_loss: 0.4712 - degree_1_loss: 0.3705 - degree_2_loss: 0.8414 - quality_loss: 0.8305 - inversion_loss: 0.9469 - root_loss: 0.7315 - key_acc 79/124 [==================>...........] - ETA: 22s - loss: 4.1916 - key_loss: 0.4713 - degree_1_loss: 0.3705 - degree_2_loss: 0.8413 - quality_loss: 0.8303 - inversion_loss: 0.9469 - root_loss: 0.7314 - key_acc 80/124 [==================>...........] - ETA: 22s - loss: 4.1911 - key_loss: 0.4713 - degree_1_loss: 0.3704 - degree_2_loss: 0.8411 - quality_loss: 0.8302 - inversion_loss: 0.9468 - root_loss: 0.7313 - key_acc 81/124 [==================>...........] - ETA: 21s - loss: 4.1905 - key_loss: 0.4713 - degree_1_loss: 0.3703 - degree_2_loss: 0.8409 - quality_loss: 0.8300 - inversion_loss: 0.9468 - root_loss: 0.7312 - key_acc 82/124 [==================>...........] - ETA: 21s - loss: 4.1898 - key_loss: 0.4713 - degree_1_loss: 0.3702 - degree_2_loss: 0.8407 - quality_loss: 0.8298 - inversion_loss: 0.9467 - root_loss: 0.7311 - key_acc 83/124 [===================>..........] - ETA: 20s - loss: 4.1893 - key_loss: 0.4714 - degree_1_loss: 0.3701 - degree_2_loss: 0.8406 - quality_loss: 0.8296 - inversion_loss: 0.9467 - root_loss: 0.7310 - key_acc 84/124 [===================>..........] - ETA: 20s - loss: 4.1887 - key_loss: 0.4714 - degree_1_loss: 0.3700 - degree_2_loss: 0.8404 - quality_loss: 0.8294 - inversion_loss: 0.9466 - root_loss: 0.7309 - key_acc 85/124 [===================>..........] - ETA: 19s - loss: 4.1881 - key_loss: 0.4714 - degree_1_loss: 0.3699 - degree_2_loss: 0.8402 - quality_loss: 0.8292 - inversion_loss: 0.9465 - root_loss: 0.7308 - key_acc 86/124 [===================>..........] - ETA: 19s - loss: 4.1875 - key_loss: 0.4714 - degree_1_loss: 0.3698 - degree_2_loss: 0.8400 - quality_loss: 0.8291 - inversion_loss: 0.9465 - root_loss: 0.7307 - key_acc 87/124 [====================>.........] - ETA: 18s - loss: 4.1869 - key_loss: 0.4715 - degree_1_loss: 0.3697 - degree_2_loss: 0.8398 - quality_loss: 0.8289 - inversion_loss: 0.9464 - root_loss: 0.7306 - key_acc 88/124 [====================>.........] - ETA: 18s - loss: 4.1863 - key_loss: 0.4715 - degree_1_loss: 0.3696 - degree_2_loss: 0.8396 - quality_loss: 0.8288 - inversion_loss: 0.9464 - root_loss: 0.7305 - key_acc 89/124 [====================>.........] - ETA: 17s - loss: 4.1857 - key_loss: 0.4715 - degree_1_loss: 0.3695 - degree_2_loss: 0.8394 - quality_loss: 0.8286 - inversion_loss: 0.9463 - root_loss: 0.7304 - key_acc 90/124 [====================>.........] - ETA: 17s - loss: 4.1852 - key_loss: 0.4715 - degree_1_loss: 0.3694 - degree_2_loss: 0.8393 - quality_loss: 0.8285 - inversion_loss: 0.9462 - root_loss: 0.7303 - key_acc 91/124 [=====================>........] - ETA: 16s - loss: 4.1847 - key_loss: 0.4715 - degree_1_loss: 0.3693 - degree_2_loss: 0.8391 - quality_loss: 0.8283 - inversion_loss: 0.9461 - root_loss: 0.7303 - key_acc 92/124 [=====================>........] - ETA: 16s - loss: 4.1841 - key_loss: 0.4715 - degree_1_loss: 0.3692 - degree_2_loss: 0.8389 - quality_loss: 0.8282 - inversion_loss: 0.9461 - root_loss: 0.7302 - key_acc 93/124 [=====================>........] - ETA: 15s - loss: 4.1836 - key_loss: 0.4715 - degree_1_loss: 0.3691 - degree_2_loss: 0.8388 - quality_loss: 0.8281 - inversion_loss: 0.9460 - root_loss: 0.7302 - key_acc 94/124 [=====================>........] - ETA: 15s - loss: 4.1832 - key_loss: 0.4715 - degree_1_loss: 0.3690 - degree_2_loss: 0.8387 - quality_loss: 0.8280 - inversion_loss: 0.9459 - root_loss: 0.7301 - key_acc 95/124 [=====================>........] - ETA: 14s - loss: 4.1827 - key_loss: 0.4715 - degree_1_loss: 0.3689 - degree_2_loss: 0.8385 - quality_loss: 0.8279 - inversion_loss: 0.9458 - root_loss: 0.7301 - key_acc 96/124 [======================>.......] - ETA: 14s - loss: 4.1822 - key_loss: 0.4715 - degree_1_loss: 0.3688 - degree_2_loss: 0.8384 - quality_loss: 0.8278 - inversion_loss: 0.9458 - root_loss: 0.7300 - key_acc 97/124 [======================>.......] - ETA: 13s - loss: 4.1818 - key_loss: 0.4715 - degree_1_loss: 0.3687 - degree_2_loss: 0.8383 - quality_loss: 0.8277 - inversion_loss: 0.9457 - root_loss: 0.7300 - key_acc 98/124 [======================>.......] - ETA: 13s - loss: 4.1813 - key_loss: 0.4714 - degree_1_loss: 0.3686 - degree_2_loss: 0.8381 - quality_loss: 0.8275 - inversion_loss: 0.9456 - root_loss: 0.7300 - key_acc 99/124 [======================>.......] - ETA: 12s - loss: 4.1807 - key_loss: 0.4714 - degree_1_loss: 0.3684 - degree_2_loss: 0.8380 - quality_loss: 0.8274 - inversion_loss: 0.9455 - root_loss: 0.7300 - key_acc100/124 [=======================>......] - ETA: 12s - loss: 4.1802 - key_loss: 0.4714 - degree_1_loss: 0.3683 - degree_2_loss: 0.8378 - quality_loss: 0.8273 - inversion_loss: 0.9454 - root_loss: 0.7299 - key_acc101/124 [=======================>......] - ETA: 11s - loss: 4.1797 - key_loss: 0.4713 - degree_1_loss: 0.3682 - degree_2_loss: 0.8377 - quality_loss: 0.8272 - inversion_loss: 0.9454 - root_loss: 0.7299 - key_acc102/124 [=======================>......] - ETA: 11s - loss: 4.1792 - key_loss: 0.4713 - degree_1_loss: 0.3681 - degree_2_loss: 0.8376 - quality_loss: 0.8271 - inversion_loss: 0.9453 - root_loss: 0.7298 - key_acc103/124 [=======================>......] - ETA: 10s - loss: 4.1788 - key_loss: 0.4713 - degree_1_loss: 0.3680 - degree_2_loss: 0.8375 - quality_loss: 0.8270 - inversion_loss: 0.9452 - root_loss: 0.7298 - key_acc104/124 [========================>.....] - ETA: 10s - loss: 4.1784 - key_loss: 0.4713 - degree_1_loss: 0.3679 - degree_2_loss: 0.8373 - quality_loss: 0.8269 - inversion_loss: 0.9451 - root_loss: 0.7298 - key_acc105/124 [========================>.....] - ETA: 9s - loss: 4.1779 - key_loss: 0.4713 - degree_1_loss: 0.3678 - degree_2_loss: 0.8372 - quality_loss: 0.8268 - inversion_loss: 0.9451 - root_loss: 0.7297 - key_accu106/124 [========================>.....] - ETA: 9s - loss: 4.1775 - key_loss: 0.4713 - degree_1_loss: 0.3677 - degree_2_loss: 0.8371 - quality_loss: 0.8267 - inversion_loss: 0.9450 - root_loss: 0.7297 - key_accu107/124 [========================>.....] - ETA: 8s - loss: 4.1771 - key_loss: 0.4713 - degree_1_loss: 0.3676 - degree_2_loss: 0.8370 - quality_loss: 0.8266 - inversion_loss: 0.9449 - root_loss: 0.7297 - key_accu108/124 [=========================>....] - ETA: 8s - loss: 4.1768 - key_loss: 0.4713 - degree_1_loss: 0.3675 - degree_2_loss: 0.8369 - quality_loss: 0.8265 - inversion_loss: 0.9449 - root_loss: 0.7296 - key_accu109/124 [=========================>....] - ETA: 7s - loss: 4.1765 - key_loss: 0.4713 - degree_1_loss: 0.3674 - degree_2_loss: 0.8369 - quality_loss: 0.8264 - inversion_loss: 0.9448 - root_loss: 0.7296 - key_accu110/124 [=========================>....] - ETA: 7s - loss: 4.1761 - key_loss: 0.4713 - degree_1_loss: 0.3673 - degree_2_loss: 0.8368 - quality_loss: 0.8263 - inversion_loss: 0.9447 - root_loss: 0.7296 - key_accu111/124 [=========================>....] - ETA: 6s - loss: 4.1758 - key_loss: 0.4713 - degree_1_loss: 0.3672 - degree_2_loss: 0.8367 - quality_loss: 0.8263 - inversion_loss: 0.9446 - root_loss: 0.7296 - key_accu112/124 [==========================>...] - ETA: 6s - loss: 4.1755 - key_loss: 0.4714 - degree_1_loss: 0.3672 - degree_2_loss: 0.8367 - quality_loss: 0.8262 - inversion_loss: 0.9446 - root_loss: 0.7296 - key_accu113/124 [==========================>...] - ETA: 5s - loss: 4.1753 - key_loss: 0.4714 - degree_1_loss: 0.3671 - degree_2_loss: 0.8366 - quality_loss: 0.8261 - inversion_loss: 0.9445 - root_loss: 0.7295 - key_accu114/124 [==========================>...] - ETA: 5s - loss: 4.1750 - key_loss: 0.4715 - degree_1_loss: 0.3670 - degree_2_loss: 0.8366 - quality_loss: 0.8260 - inversion_loss: 0.9444 - root_loss: 0.7295 - key_accu115/124 [==========================>...] - ETA: 4s - loss: 4.1747 - key_loss: 0.4715 - degree_1_loss: 0.3669 - degree_2_loss: 0.8365 - quality_loss: 0.8260 - inversion_loss: 0.9444 - root_loss: 0.7295 - key_accu116/124 [===========================>..] - ETA: 4s - loss: 4.1744 - key_loss: 0.4716 - degree_1_loss: 0.3668 - degree_2_loss: 0.8365 - quality_loss: 0.8259 - inversion_loss: 0.9443 - root_loss: 0.7294 - key_accu117/124 [===========================>..] - ETA: 3s - loss: 4.1742 - key_loss: 0.4716 - degree_1_loss: 0.3667 - degree_2_loss: 0.8364 - quality_loss: 0.8258 - inversion_loss: 0.9442 - root_loss: 0.7294 - key_accu118/124 [===========================>..] - ETA: 3s - loss: 4.1739 - key_loss: 0.4717 - degree_1_loss: 0.3666 - degree_2_loss: 0.8364 - quality_loss: 0.8257 - inversion_loss: 0.9442 - root_loss: 0.7293 - key_accu119/124 [===========================>..] - ETA: 2s - loss: 4.1737 - key_loss: 0.4718 - degree_1_loss: 0.3666 - degree_2_loss: 0.8363 - quality_loss: 0.8256 - inversion_loss: 0.9441 - root_loss: 0.7293 - key_accu120/124 [============================>.] - ETA: 2s - loss: 4.1735 - key_loss: 0.4718 - degree_1_loss: 0.3665 - degree_2_loss: 0.8363 - quality_loss: 0.8255 - inversion_loss: 0.9440 - root_loss: 0.7293 - key_accu121/124 [============================>.] - ETA: 1s - loss: 4.1732 - key_loss: 0.4719 - degree_1_loss: 0.3664 - degree_2_loss: 0.8363 - quality_loss: 0.8255 - inversion_loss: 0.9440 - root_loss: 0.7292 - key_accu122/124 [============================>.] - ETA: 1s - loss: 4.1730 - key_loss: 0.4720 - degree_1_loss: 0.3663 - degree_2_loss: 0.8362 - quality_loss: 0.8254 - inversion_loss: 0.9439 - root_loss: 0.7292 - key_accu123/124 [============================>.] - ETA: 0s - loss: 4.1727 - key_loss: 0.4720 - degree_1_loss: 0.3662 - degree_2_loss: 0.8362 - quality_loss: 0.8253 - inversion_loss: 0.9439 - root_loss: 0.7291 - key_accu124/124 [==============================] - ETA: 0s - loss: 4.1724 - key_loss: 0.4721 - degree_1_loss: 0.3662 - degree_2_loss: 0.8361 - quality_loss: 0.8252 - inversion_loss: 0.9438 - root_loss: 0.7291 - key_accu124/124 [==============================] - 64s 513ms/step - loss: 4.1721 - key_loss: 0.4721 - degree_1_loss: 0.3661 - degree_2_loss: 0.8360 - quality_loss: 0.8251 - inversion_loss: 0.9437 - root_loss: 0.7290 - key_accuracy: 0.8431 - degree_1_accuracy: 0.9021 - degree_2_accuracy: 0.6869 - quality_accuracy: 0.6975 - inversion_accuracy: 0.5912 - root_accuracy: 0.7500 - val_loss: 4.4118 - val_key_loss: 0.6642 - val_degree_1_loss: 0.3561 - val_degree_2_loss: 0.9122 - val_quality_loss: 0.8398 - val_inversion_loss: 0.9198 - val_root_loss: 0.7198 - val_key_accuracy: 0.7827 - val_degree_1_accuracy: 0.9097 - val_degree_2_accuracy: 0.6649 - val_quality_accuracy: 0.7028 - val_inversion_accuracy: 0.6237 - val_root_accuracy: 0.7687

loss: 4.1721
key_loss: 0.4721
degree_1_loss: 0.3661
degree_2_loss: 0.8360
quality_loss: 0.8251
inversion_loss: 0.9437
root_loss: 0.7290

key_accuracy: 0.8431
degree_1_accuracy: 0.9021
degree_2_accuracy: 0.6869
quality_accuracy: 0.6975
inversion_accuracy: 0.5912
root_accuracy: 0.7500

val_loss: 4.4118
val_key_loss: 0.6642
val_degree_1_loss: 0.3561
val_degree_2_loss: 0.9122
val_quality_loss: 0.8398
val_inversion_loss: 0.9198
val_root_loss: 0.7198

val_key_accuracy: 0.7827
val_degree_1_accuracy: 0.9097
val_degree_2_accuracy: 0.6649
val_quality_accuracy: 0.7028
val_inversion_accuracy: 0.6237
val_root_accuracy: 0.7687

They seem pretty similar.

Now with N_OCTAVES==5.

Epoch 28/100
  1/124 [..............................] - ETA: 1:21 - loss: 3.6382 - key_loss: 0.3310 - degree_1_loss: 0.3306 - degree_2_loss: 0.7435 - quality_loss: 0.6863 - inversion_loss: 0.9024 - root_loss: 0.6444 - key_ac  2/124 [..............................] - ETA: 1:04 - loss: 3.7023 - key_loss: 0.3477 - degree_1_loss: 0.3494 - degree_2_loss: 0.7439 - quality_loss: 0.6983 - inversion_loss: 0.9134 - root_loss: 0.6497 - key_ac  3/124 [..............................] - ETA: 1:02 - loss: 3.7399 - key_loss: 0.3550 - degree_1_loss: 0.3456 - degree_2_loss: 0.7461 - quality_loss: 0.7204 - inversion_loss: 0.9177 - root_loss: 0.6552 - key_ac  4/124 [..............................] - ETA: 1:01 - loss: 3.7797 - key_loss: 0.3643 - degree_1_loss: 0.3404 - degree_2_loss: 0.7527 - quality_loss: 0.7410 - inversion_loss: 0.9190 - root_loss: 0.6622 - key_ac  5/124 [>.............................] - ETA: 1:00 - loss: 3.8100 - key_loss: 0.3704 - degree_1_loss: 0.3409 - degree_2_loss: 0.7588 - quality_loss: 0.7514 - inversion_loss: 0.9205 - root_loss: 0.6681 - key_ac  6/124 [>.............................] - ETA: 1:00 - loss: 3.8417 - key_loss: 0.3799 - degree_1_loss: 0.3447 - degree_2_loss: 0.7633 - quality_loss: 0.7584 - inversion_loss: 0.9212 - root_loss: 0.6742 - key_ac  7/124 [>.............................] - ETA: 59s - loss: 3.8560 - key_loss: 0.3826 - degree_1_loss: 0.3473 - degree_2_loss: 0.7660 - quality_loss: 0.7630 - inversion_loss: 0.9196 - root_loss: 0.6775 - key_acc  8/124 [>.............................] - ETA: 59s - loss: 3.8584 - key_loss: 0.3828 - degree_1_loss: 0.3478 - degree_2_loss: 0.7665 - quality_loss: 0.7650 - inversion_loss: 0.9168 - root_loss: 0.6796 - key_acc  9/124 [=>............................] - ETA: 58s - loss: 3.8645 - key_loss: 0.3838 - degree_1_loss: 0.3487 - degree_2_loss: 0.7667 - quality_loss: 0.7674 - inversion_loss: 0.9154 - root_loss: 0.6825 - key_acc 10/124 [=>............................] - ETA: 58s - loss: 3.8716 - key_loss: 0.3851 - degree_1_loss: 0.3493 - degree_2_loss: 0.7675 - quality_loss: 0.7699 - inversion_loss: 0.9152 - root_loss: 0.6847 - key_acc 11/124 [=>............................] - ETA: 57s - loss: 3.8783 - key_loss: 0.3865 - degree_1_loss: 0.3501 - degree_2_loss: 0.7685 - quality_loss: 0.7724 - inversion_loss: 0.9146 - root_loss: 0.6863 - key_acc 12/124 [=>............................] - ETA: 57s - loss: 3.8777 - key_loss: 0.3863 - degree_1_loss: 0.3499 - degree_2_loss: 0.7684 - quality_loss: 0.7734 - inversion_loss: 0.9131 - root_loss: 0.6866 - key_acc 13/124 [==>...........................] - ETA: 57s - loss: 3.8781 - key_loss: 0.3863 - degree_1_loss: 0.3499 - degree_2_loss: 0.7684 - quality_loss: 0.7747 - inversion_loss: 0.9121 - root_loss: 0.6867 - key_acc 14/124 [==>...........................] - ETA: 56s - loss: 3.8797 - key_loss: 0.3869 - degree_1_loss: 0.3501 - degree_2_loss: 0.7683 - quality_loss: 0.7762 - inversion_loss: 0.9118 - root_loss: 0.6863 - key_acc 15/124 [==>...........................] - ETA: 55s - loss: 3.8826 - key_loss: 0.3882 - degree_1_loss: 0.3503 - degree_2_loss: 0.7681 - quality_loss: 0.7777 - inversion_loss: 0.9124 - root_loss: 0.6859 - key_acc 16/124 [==>...........................] - ETA: 55s - loss: 3.8824 - key_loss: 0.3889 - degree_1_loss: 0.3500 - degree_2_loss: 0.7678 - quality_loss: 0.7787 - inversion_loss: 0.9122 - root_loss: 0.6849 - key_acc 17/124 [===>..........................] - ETA: 54s - loss: 3.8821 - key_loss: 0.3894 - degree_1_loss: 0.3500 - degree_2_loss: 0.7671 - quality_loss: 0.7794 - inversion_loss: 0.9122 - root_loss: 0.6840 - key_acc 18/124 [===>..........................] - ETA: 54s - loss: 3.8814 - key_loss: 0.3897 - degree_1_loss: 0.3497 - degree_2_loss: 0.7665 - quality_loss: 0.7800 - inversion_loss: 0.9123 - root_loss: 0.6831 - key_acc 19/124 [===>..........................] - ETA: 53s - loss: 3.8820 - key_loss: 0.3903 - degree_1_loss: 0.3496 - degree_2_loss: 0.7663 - quality_loss: 0.7809 - inversion_loss: 0.9123 - root_loss: 0.6826 - key_acc 20/124 [===>..........................] - ETA: 53s - loss: 3.8834 - key_loss: 0.3908 - degree_1_loss: 0.3495 - degree_2_loss: 0.7664 - quality_loss: 0.7819 - inversion_loss: 0.9125 - root_loss: 0.6823 - key_acc 21/124 [====>.........................] - ETA: 52s - loss: 3.8834 - key_loss: 0.3912 - degree_1_loss: 0.3493 - degree_2_loss: 0.7662 - quality_loss: 0.7825 - inversion_loss: 0.9126 - root_loss: 0.6816 - key_acc 22/124 [====>.........................] - ETA: 52s - loss: 3.8826 - key_loss: 0.3916 - degree_1_loss: 0.3490 - degree_2_loss: 0.7657 - quality_loss: 0.7828 - inversion_loss: 0.9126 - root_loss: 0.6809 - key_acc 23/124 [====>.........................] - ETA: 51s - loss: 3.8825 - key_loss: 0.3919 - degree_1_loss: 0.3489 - degree_2_loss: 0.7654 - quality_loss: 0.7833 - inversion_loss: 0.9126 - root_loss: 0.6805 - key_acc 24/124 [====>.........................] - ETA: 51s - loss: 3.8825 - key_loss: 0.3920 - degree_1_loss: 0.3487 - degree_2_loss: 0.7653 - quality_loss: 0.7838 - inversion_loss: 0.9128 - root_loss: 0.6800 - key_acc 25/124 [=====>........................] - ETA: 50s - loss: 3.8828 - key_loss: 0.3920 - degree_1_loss: 0.3484 - degree_2_loss: 0.7653 - quality_loss: 0.7844 - inversion_loss: 0.9129 - root_loss: 0.6797 - key_acc 26/124 [=====>........................] - ETA: 50s - loss: 3.8826 - key_loss: 0.3921 - degree_1_loss: 0.3482 - degree_2_loss: 0.7653 - quality_loss: 0.7847 - inversion_loss: 0.9130 - root_loss: 0.6792 - key_acc 27/124 [=====>........................] - ETA: 49s - loss: 3.8822 - key_loss: 0.3922 - degree_1_loss: 0.3480 - degree_2_loss: 0.7651 - quality_loss: 0.7850 - inversion_loss: 0.9131 - root_loss: 0.6788 - key_acc 28/124 [=====>........................] - ETA: 49s - loss: 3.8814 - key_loss: 0.3923 - degree_1_loss: 0.3477 - degree_2_loss: 0.7648 - quality_loss: 0.7851 - inversion_loss: 0.9132 - root_loss: 0.6782 - key_acc 29/124 [======>.......................] - ETA: 48s - loss: 3.8815 - key_loss: 0.3926 - degree_1_loss: 0.3475 - degree_2_loss: 0.7647 - quality_loss: 0.7854 - inversion_loss: 0.9134 - root_loss: 0.6779 - key_acc 30/124 [======>.......................] - ETA: 48s - loss: 3.8818 - key_loss: 0.3929 - degree_1_loss: 0.3474 - degree_2_loss: 0.7647 - quality_loss: 0.7857 - inversion_loss: 0.9135 - root_loss: 0.6776 - key_acc 31/124 [======>.......................] - ETA: 48s - loss: 3.8816 - key_loss: 0.3931 - degree_1_loss: 0.3472 - degree_2_loss: 0.7646 - quality_loss: 0.7859 - inversion_loss: 0.9135 - root_loss: 0.6772 - key_acc 32/124 [======>.......................] - ETA: 47s - loss: 3.8810 - key_loss: 0.3933 - degree_1_loss: 0.3469 - degree_2_loss: 0.7644 - quality_loss: 0.7860 - inversion_loss: 0.9136 - root_loss: 0.6768 - key_acc 33/124 [======>.......................] - ETA: 47s - loss: 3.8811 - key_loss: 0.3936 - degree_1_loss: 0.3467 - degree_2_loss: 0.7643 - quality_loss: 0.7862 - inversion_loss: 0.9137 - root_loss: 0.6766 - key_acc 34/124 [=======>......................] - ETA: 47s - loss: 3.8814 - key_loss: 0.3939 - degree_1_loss: 0.3465 - degree_2_loss: 0.7643 - quality_loss: 0.7864 - inversion_loss: 0.9139 - root_loss: 0.6763 - key_acc 35/124 [=======>......................] - ETA: 46s - loss: 3.8817 - key_loss: 0.3943 - degree_1_loss: 0.3464 - degree_2_loss: 0.7643 - quality_loss: 0.7866 - inversion_loss: 0.9141 - root_loss: 0.6761 - key_acc 36/124 [=======>......................] - ETA: 46s - loss: 3.8820 - key_loss: 0.3947 - degree_1_loss: 0.3463 - degree_2_loss: 0.7642 - quality_loss: 0.7867 - inversion_loss: 0.9142 - root_loss: 0.6759 - key_acc 37/124 [=======>......................] - ETA: 45s - loss: 3.8821 - key_loss: 0.3950 - degree_1_loss: 0.3461 - degree_2_loss: 0.7642 - quality_loss: 0.7868 - inversion_loss: 0.9143 - root_loss: 0.6757 - key_acc 38/124 [========>.....................] - ETA: 45s - loss: 3.8817 - key_loss: 0.3952 - degree_1_loss: 0.3459 - degree_2_loss: 0.7640 - quality_loss: 0.7867 - inversion_loss: 0.9144 - root_loss: 0.6754 - key_acc 39/124 [========>.....................] - ETA: 44s - loss: 3.8814 - key_loss: 0.3954 - degree_1_loss: 0.3458 - degree_2_loss: 0.7639 - quality_loss: 0.7867 - inversion_loss: 0.9144 - root_loss: 0.6751 - key_acc 40/124 [========>.....................] - ETA: 44s - loss: 3.8805 - key_loss: 0.3955 - degree_1_loss: 0.3456 - degree_2_loss: 0.7637 - quality_loss: 0.7866 - inversion_loss: 0.9144 - root_loss: 0.6748 - key_acc 41/124 [========>.....................] - ETA: 43s - loss: 3.8796 - key_loss: 0.3955 - degree_1_loss: 0.3454 - degree_2_loss: 0.7635 - quality_loss: 0.7865 - inversion_loss: 0.9144 - root_loss: 0.6745 - key_acc 42/124 [=========>....................] - ETA: 43s - loss: 3.8788 - key_loss: 0.3955 - degree_1_loss: 0.3452 - degree_2_loss: 0.7633 - quality_loss: 0.7863 - inversion_loss: 0.9144 - root_loss: 0.6742 - key_acc 43/124 [=========>....................] - ETA: 42s - loss: 3.8780 - key_loss: 0.3954 - degree_1_loss: 0.3450 - degree_2_loss: 0.7631 - quality_loss: 0.7862 - inversion_loss: 0.9145 - root_loss: 0.6739 - key_acc 44/124 [=========>....................] - ETA: 41s - loss: 3.8771 - key_loss: 0.3954 - degree_1_loss: 0.3448 - degree_2_loss: 0.7628 - quality_loss: 0.7860 - inversion_loss: 0.9145 - root_loss: 0.6736 - key_acc 45/124 [=========>....................] - ETA: 41s - loss: 3.8764 - key_loss: 0.3953 - degree_1_loss: 0.3447 - degree_2_loss: 0.7627 - quality_loss: 0.7859 - inversion_loss: 0.9145 - root_loss: 0.6733 - key_acc 46/124 [==========>...................] - ETA: 40s - loss: 3.8757 - key_loss: 0.3952 - degree_1_loss: 0.3446 - degree_2_loss: 0.7625 - quality_loss: 0.7858 - inversion_loss: 0.9145 - root_loss: 0.6731 - key_acc 47/124 [==========>...................] - ETA: 40s - loss: 3.8752 - key_loss: 0.3951 - degree_1_loss: 0.3446 - degree_2_loss: 0.7623 - quality_loss: 0.7857 - inversion_loss: 0.9145 - root_loss: 0.6729 - key_acc 48/124 [==========>...................] - ETA: 39s - loss: 3.8749 - key_loss: 0.3951 - degree_1_loss: 0.3446 - degree_2_loss: 0.7623 - quality_loss: 0.7857 - inversion_loss: 0.9145 - root_loss: 0.6728 - key_acc 49/124 [==========>...................] - ETA: 39s - loss: 3.8750 - key_loss: 0.3951 - degree_1_loss: 0.3446 - degree_2_loss: 0.7622 - quality_loss: 0.7857 - inversion_loss: 0.9146 - root_loss: 0.6727 - key_acc 50/124 [===========>..................] - ETA: 38s - loss: 3.8751 - key_loss: 0.3952 - degree_1_loss: 0.3446 - degree_2_loss: 0.7622 - quality_loss: 0.7857 - inversion_loss: 0.9147 - root_loss: 0.6727 - key_acc 51/124 [===========>..................] - ETA: 38s - loss: 3.8750 - key_loss: 0.3952 - degree_1_loss: 0.3446 - degree_2_loss: 0.7622 - quality_loss: 0.7857 - inversion_loss: 0.9147 - root_loss: 0.6726 - key_acc 52/124 [===========>..................] - ETA: 37s - loss: 3.8751 - key_loss: 0.3952 - degree_1_loss: 0.3446 - degree_2_loss: 0.7622 - quality_loss: 0.7858 - inversion_loss: 0.9148 - root_loss: 0.6726 - key_acc 53/124 [===========>..................] - ETA: 37s - loss: 3.8752 - key_loss: 0.3952 - degree_1_loss: 0.3446 - degree_2_loss: 0.7622 - quality_loss: 0.7859 - inversion_loss: 0.9149 - root_loss: 0.6725 - key_acc 54/124 [============>.................] - ETA: 36s - loss: 3.8752 - key_loss: 0.3952 - degree_1_loss: 0.3445 - degree_2_loss: 0.7621 - quality_loss: 0.7859 - inversion_loss: 0.9149 - root_loss: 0.6725 - key_acc 55/124 [============>.................] - ETA: 36s - loss: 3.8752 - key_loss: 0.3952 - degree_1_loss: 0.3445 - degree_2_loss: 0.7621 - quality_loss: 0.7859 - inversion_loss: 0.9150 - root_loss: 0.6724 - key_acc 56/124 [============>.................] - ETA: 35s - loss: 3.8751 - key_loss: 0.3953 - degree_1_loss: 0.3444 - degree_2_loss: 0.7621 - quality_loss: 0.7860 - inversion_loss: 0.9150 - root_loss: 0.6723 - key_acc 57/124 [============>.................] - ETA: 35s - loss: 3.8752 - key_loss: 0.3953 - degree_1_loss: 0.3445 - degree_2_loss: 0.7621 - quality_loss: 0.7860 - inversion_loss: 0.9151 - root_loss: 0.6723 - key_acc 58/124 [=============>................] - ETA: 34s - loss: 3.8750 - key_loss: 0.3953 - degree_1_loss: 0.3444 - degree_2_loss: 0.7620 - quality_loss: 0.7860 - inversion_loss: 0.9151 - root_loss: 0.6722 - key_acc 59/124 [=============>................] - ETA: 34s - loss: 3.8747 - key_loss: 0.3953 - degree_1_loss: 0.3444 - degree_2_loss: 0.7619 - quality_loss: 0.7860 - inversion_loss: 0.9151 - root_loss: 0.6721 - key_acc 60/124 [=============>................] - ETA: 33s - loss: 3.8745 - key_loss: 0.3952 - degree_1_loss: 0.3444 - degree_2_loss: 0.7618 - quality_loss: 0.7859 - inversion_loss: 0.9152 - root_loss: 0.6720 - key_acc 61/124 [=============>................] - ETA: 33s - loss: 3.8742 - key_loss: 0.3951 - degree_1_loss: 0.3444 - degree_2_loss: 0.7617 - quality_loss: 0.7859 - inversion_loss: 0.9152 - root_loss: 0.6719 - key_acc 62/124 [==============>...............] - ETA: 32s - loss: 3.8737 - key_loss: 0.3951 - degree_1_loss: 0.3443 - degree_2_loss: 0.7615 - quality_loss: 0.7858 - inversion_loss: 0.9152 - root_loss: 0.6718 - key_acc 63/124 [==============>...............] - ETA: 32s - loss: 3.8734 - key_loss: 0.3950 - degree_1_loss: 0.3443 - degree_2_loss: 0.7614 - quality_loss: 0.7857 - inversion_loss: 0.9152 - root_loss: 0.6718 - key_acc 64/124 [==============>...............] - ETA: 31s - loss: 3.8732 - key_loss: 0.3950 - degree_1_loss: 0.3443 - degree_2_loss: 0.7613 - quality_loss: 0.7857 - inversion_loss: 0.9153 - root_loss: 0.6717 - key_acc 65/124 [==============>...............] - ETA: 31s - loss: 3.8730 - key_loss: 0.3949 - degree_1_loss: 0.3443 - degree_2_loss: 0.7611 - quality_loss: 0.7857 - inversion_loss: 0.9153 - root_loss: 0.6717 - key_acc 66/124 [==============>...............] - ETA: 30s - loss: 3.8728 - key_loss: 0.3949 - degree_1_loss: 0.3443 - degree_2_loss: 0.7610 - quality_loss: 0.7856 - inversion_loss: 0.9154 - root_loss: 0.6716 - key_acc 67/124 [===============>..............] - ETA: 29s - loss: 3.8726 - key_loss: 0.3948 - degree_1_loss: 0.3443 - degree_2_loss: 0.7609 - quality_loss: 0.7856 - inversion_loss: 0.9154 - root_loss: 0.6716 - key_acc 68/124 [===============>..............] - ETA: 29s - loss: 3.8725 - key_loss: 0.3947 - degree_1_loss: 0.3442 - degree_2_loss: 0.7608 - quality_loss: 0.7856 - inversion_loss: 0.9154 - root_loss: 0.6716 - key_acc 69/124 [===============>..............] - ETA: 28s - loss: 3.8723 - key_loss: 0.3947 - degree_1_loss: 0.3442 - degree_2_loss: 0.7608 - quality_loss: 0.7856 - inversion_loss: 0.9155 - root_loss: 0.6716 - key_acc 70/124 [===============>..............] - ETA: 28s - loss: 3.8723 - key_loss: 0.3946 - degree_1_loss: 0.3442 - degree_2_loss: 0.7607 - quality_loss: 0.7856 - inversion_loss: 0.9155 - root_loss: 0.6716 - key_acc 71/124 [================>.............] - ETA: 27s - loss: 3.8722 - key_loss: 0.3946 - degree_1_loss: 0.3442 - degree_2_loss: 0.7607 - quality_loss: 0.7856 - inversion_loss: 0.9156 - root_loss: 0.6716 - key_acc 72/124 [================>.............] - ETA: 27s - loss: 3.8721 - key_loss: 0.3945 - degree_1_loss: 0.3442 - degree_2_loss: 0.7607 - quality_loss: 0.7856 - inversion_loss: 0.9156 - root_loss: 0.6716 - key_acc 73/124 [================>.............] - ETA: 26s - loss: 3.8721 - key_loss: 0.3944 - degree_1_loss: 0.3442 - degree_2_loss: 0.7606 - quality_loss: 0.7856 - inversion_loss: 0.9157 - root_loss: 0.6717 - key_acc 74/124 [================>.............] - ETA: 26s - loss: 3.8720 - key_loss: 0.3943 - degree_1_loss: 0.3442 - degree_2_loss: 0.7606 - quality_loss: 0.7855 - inversion_loss: 0.9157 - root_loss: 0.6717 - key_acc 75/124 [=================>............] - ETA: 25s - loss: 3.8718 - key_loss: 0.3942 - degree_1_loss: 0.3442 - degree_2_loss: 0.7605 - quality_loss: 0.7855 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 76/124 [=================>............] - ETA: 25s - loss: 3.8716 - key_loss: 0.3941 - degree_1_loss: 0.3442 - degree_2_loss: 0.7605 - quality_loss: 0.7855 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 77/124 [=================>............] - ETA: 24s - loss: 3.8715 - key_loss: 0.3940 - degree_1_loss: 0.3442 - degree_2_loss: 0.7604 - quality_loss: 0.7855 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 78/124 [=================>............] - ETA: 24s - loss: 3.8714 - key_loss: 0.3938 - degree_1_loss: 0.3442 - degree_2_loss: 0.7604 - quality_loss: 0.7855 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 79/124 [==================>...........] - ETA: 23s - loss: 3.8711 - key_loss: 0.3937 - degree_1_loss: 0.3442 - degree_2_loss: 0.7603 - quality_loss: 0.7854 - inversion_loss: 0.9159 - root_loss: 0.6716 - key_acc 80/124 [==================>...........] - ETA: 23s - loss: 3.8710 - key_loss: 0.3936 - degree_1_loss: 0.3442 - degree_2_loss: 0.7602 - quality_loss: 0.7854 - inversion_loss: 0.9159 - root_loss: 0.6716 - key_acc 81/124 [==================>...........] - ETA: 22s - loss: 3.8708 - key_loss: 0.3935 - degree_1_loss: 0.3442 - degree_2_loss: 0.7602 - quality_loss: 0.7854 - inversion_loss: 0.9160 - root_loss: 0.6716 - key_acc 82/124 [==================>...........] - ETA: 22s - loss: 3.8706 - key_loss: 0.3934 - degree_1_loss: 0.3442 - degree_2_loss: 0.7601 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6716 - key_acc 83/124 [===================>..........] - ETA: 21s - loss: 3.8704 - key_loss: 0.3933 - degree_1_loss: 0.3441 - degree_2_loss: 0.7601 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6716 - key_acc 84/124 [===================>..........] - ETA: 20s - loss: 3.8702 - key_loss: 0.3932 - degree_1_loss: 0.3442 - degree_2_loss: 0.7600 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6716 - key_acc 85/124 [===================>..........] - ETA: 20s - loss: 3.8702 - key_loss: 0.3931 - degree_1_loss: 0.3442 - degree_2_loss: 0.7600 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6716 - key_acc 86/124 [===================>..........] - ETA: 19s - loss: 3.8700 - key_loss: 0.3929 - degree_1_loss: 0.3442 - degree_2_loss: 0.7599 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6717 - key_acc 87/124 [====================>.........] - ETA: 19s - loss: 3.8698 - key_loss: 0.3928 - degree_1_loss: 0.3442 - degree_2_loss: 0.7599 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6717 - key_acc 88/124 [====================>.........] - ETA: 18s - loss: 3.8696 - key_loss: 0.3927 - degree_1_loss: 0.3442 - degree_2_loss: 0.7598 - quality_loss: 0.7853 - inversion_loss: 0.9160 - root_loss: 0.6716 - key_acc 89/124 [====================>.........] - ETA: 18s - loss: 3.8694 - key_loss: 0.3926 - degree_1_loss: 0.3442 - degree_2_loss: 0.7598 - quality_loss: 0.7852 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 90/124 [====================>.........] - ETA: 17s - loss: 3.8691 - key_loss: 0.3924 - degree_1_loss: 0.3442 - degree_2_loss: 0.7597 - quality_loss: 0.7852 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 91/124 [=====================>........] - ETA: 17s - loss: 3.8689 - key_loss: 0.3923 - degree_1_loss: 0.3442 - degree_2_loss: 0.7597 - quality_loss: 0.7852 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 92/124 [=====================>........] - ETA: 16s - loss: 3.8687 - key_loss: 0.3922 - degree_1_loss: 0.3442 - degree_2_loss: 0.7596 - quality_loss: 0.7852 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 93/124 [=====================>........] - ETA: 16s - loss: 3.8685 - key_loss: 0.3921 - degree_1_loss: 0.3442 - degree_2_loss: 0.7596 - quality_loss: 0.7852 - inversion_loss: 0.9159 - root_loss: 0.6717 - key_acc 94/124 [=====================>........] - ETA: 15s - loss: 3.8684 - key_loss: 0.3920 - degree_1_loss: 0.3441 - degree_2_loss: 0.7595 - quality_loss: 0.7851 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 95/124 [=====================>........] - ETA: 15s - loss: 3.8682 - key_loss: 0.3920 - degree_1_loss: 0.3441 - degree_2_loss: 0.7595 - quality_loss: 0.7851 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 96/124 [======================>.......] - ETA: 14s - loss: 3.8681 - key_loss: 0.3920 - degree_1_loss: 0.3441 - degree_2_loss: 0.7595 - quality_loss: 0.7851 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 97/124 [======================>.......] - ETA: 14s - loss: 3.8679 - key_loss: 0.3919 - degree_1_loss: 0.3441 - degree_2_loss: 0.7594 - quality_loss: 0.7851 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 98/124 [======================>.......] - ETA: 13s - loss: 3.8678 - key_loss: 0.3918 - degree_1_loss: 0.3441 - degree_2_loss: 0.7594 - quality_loss: 0.7850 - inversion_loss: 0.9158 - root_loss: 0.6717 - key_acc 99/124 [======================>.......] - ETA: 13s - loss: 3.8676 - key_loss: 0.3918 - degree_1_loss: 0.3441 - degree_2_loss: 0.7593 - quality_loss: 0.7850 - inversion_loss: 0.9158 - root_loss: 0.6716 - key_acc100/124 [=======================>......] - ETA: 12s - loss: 3.8674 - key_loss: 0.3917 - degree_1_loss: 0.3440 - degree_2_loss: 0.7593 - quality_loss: 0.7850 - inversion_loss: 0.9158 - root_loss: 0.6716 - key_acc101/124 [=======================>......] - ETA: 12s - loss: 3.8672 - key_loss: 0.3916 - degree_1_loss: 0.3440 - degree_2_loss: 0.7592 - quality_loss: 0.7849 - inversion_loss: 0.9158 - root_loss: 0.6716 - key_acc102/124 [=======================>......] - ETA: 11s - loss: 3.8670 - key_loss: 0.3916 - degree_1_loss: 0.3440 - degree_2_loss: 0.7592 - quality_loss: 0.7849 - inversion_loss: 0.9158 - root_loss: 0.6716 - key_acc103/124 [=======================>......] - ETA: 11s - loss: 3.8669 - key_loss: 0.3915 - degree_1_loss: 0.3440 - degree_2_loss: 0.7592 - quality_loss: 0.7848 - inversion_loss: 0.9157 - root_loss: 0.6716 - key_acc104/124 [========================>.....] - ETA: 10s - loss: 3.8667 - key_loss: 0.3915 - degree_1_loss: 0.3440 - degree_2_loss: 0.7591 - quality_loss: 0.7848 - inversion_loss: 0.9157 - root_loss: 0.6716 - key_acc105/124 [========================>.....] - ETA: 9s - loss: 3.8665 - key_loss: 0.3914 - degree_1_loss: 0.3440 - degree_2_loss: 0.7591 - quality_loss: 0.7848 - inversion_loss: 0.9157 - root_loss: 0.6716 - key_accu106/124 [========================>.....] - ETA: 9s - loss: 3.8664 - key_loss: 0.3913 - degree_1_loss: 0.3440 - degree_2_loss: 0.7591 - quality_loss: 0.7847 - inversion_loss: 0.9156 - root_loss: 0.6716 - key_accu107/124 [========================>.....] - ETA: 8s - loss: 3.8662 - key_loss: 0.3913 - degree_1_loss: 0.3440 - degree_2_loss: 0.7590 - quality_loss: 0.7847 - inversion_loss: 0.9156 - root_loss: 0.6716 - key_accu108/124 [=========================>....] - ETA: 8s - loss: 3.8660 - key_loss: 0.3912 - degree_1_loss: 0.3440 - degree_2_loss: 0.7590 - quality_loss: 0.7847 - inversion_loss: 0.9155 - root_loss: 0.6717 - key_accu109/124 [=========================>....] - ETA: 7s - loss: 3.8658 - key_loss: 0.3912 - degree_1_loss: 0.3439 - degree_2_loss: 0.7590 - quality_loss: 0.7846 - inversion_loss: 0.9155 - root_loss: 0.6717 - key_accu110/124 [=========================>....] - ETA: 7s - loss: 3.8656 - key_loss: 0.3911 - degree_1_loss: 0.3439 - degree_2_loss: 0.7589 - quality_loss: 0.7846 - inversion_loss: 0.9154 - root_loss: 0.6717 - key_accu111/124 [=========================>....] - ETA: 6s - loss: 3.8655 - key_loss: 0.3911 - degree_1_loss: 0.3439 - degree_2_loss: 0.7589 - quality_loss: 0.7845 - inversion_loss: 0.9154 - root_loss: 0.6717 - key_accu112/124 [==========================>...] - ETA: 6s - loss: 3.8653 - key_loss: 0.3911 - degree_1_loss: 0.3439 - degree_2_loss: 0.7588 - quality_loss: 0.7845 - inversion_loss: 0.9153 - root_loss: 0.6717 - key_accu113/124 [==========================>...] - ETA: 5s - loss: 3.8652 - key_loss: 0.3911 - degree_1_loss: 0.3439 - degree_2_loss: 0.7588 - quality_loss: 0.7844 - inversion_loss: 0.9152 - root_loss: 0.6717 - key_accu114/124 [==========================>...] - ETA: 5s - loss: 3.8651 - key_loss: 0.3910 - degree_1_loss: 0.3439 - degree_2_loss: 0.7588 - quality_loss: 0.7844 - inversion_loss: 0.9152 - root_loss: 0.6718 - key_accu115/124 [==========================>...] - ETA: 4s - loss: 3.8650 - key_loss: 0.3910 - degree_1_loss: 0.3439 - degree_2_loss: 0.7587 - quality_loss: 0.7844 - inversion_loss: 0.9151 - root_loss: 0.6718 - key_accu116/124 [===========================>..] - ETA: 4s - loss: 3.8649 - key_loss: 0.3910 - degree_1_loss: 0.3439 - degree_2_loss: 0.7587 - quality_loss: 0.7844 - inversion_loss: 0.9151 - root_loss: 0.6718 - key_accu117/124 [===========================>..] - ETA: 3s - loss: 3.8649 - key_loss: 0.3910 - degree_1_loss: 0.3439 - degree_2_loss: 0.7587 - quality_loss: 0.7843 - inversion_loss: 0.9150 - root_loss: 0.6719 - key_accu118/124 [===========================>..] - ETA: 3s - loss: 3.8648 - key_loss: 0.3910 - degree_1_loss: 0.3439 - degree_2_loss: 0.7587 - quality_loss: 0.7843 - inversion_loss: 0.9150 - root_loss: 0.6719 - key_accu119/124 [===========================>..] - ETA: 2s - loss: 3.8647 - key_loss: 0.3909 - degree_1_loss: 0.3439 - degree_2_loss: 0.7587 - quality_loss: 0.7843 - inversion_loss: 0.9149 - root_loss: 0.6719 - key_accu120/124 [============================>.] - ETA: 2s - loss: 3.8646 - key_loss: 0.3909 - degree_1_loss: 0.3439 - degree_2_loss: 0.7587 - quality_loss: 0.7843 - inversion_loss: 0.9149 - root_loss: 0.6720 - key_accu121/124 [============================>.] - ETA: 1s - loss: 3.8644 - key_loss: 0.3909 - degree_1_loss: 0.3439 - degree_2_loss: 0.7586 - quality_loss: 0.7842 - inversion_loss: 0.9148 - root_loss: 0.6720 - key_accu122/124 [============================>.] - ETA: 1s - loss: 3.8643 - key_loss: 0.3909 - degree_1_loss: 0.3439 - degree_2_loss: 0.7586 - quality_loss: 0.7842 - inversion_loss: 0.9147 - root_loss: 0.6720 - key_accu123/124 [============================>.] - ETA: 0s - loss: 3.8641 - key_loss: 0.3909 - degree_1_loss: 0.3439 - degree_2_loss: 0.7586 - quality_loss: 0.7841 - inversion_loss: 0.9147 - root_loss: 0.6721 - key_accu124/124 [==============================] - ETA: 0s - loss: 3.8640 - key_loss: 0.3908 - degree_1_loss: 0.3438 - degree_2_loss: 0.7585 - quality_loss: 0.7841 - inversion_loss: 0.9146 - root_loss: 0.6721 - key_accu124/124 [==============================] - 66s 530ms/step - loss: 3.8639 - key_loss: 0.3908 - degree_1_loss: 0.3438 - degree_2_loss: 0.7585 - quality_loss: 0.7840 - inversion_loss: 0.9146 - root_loss: 0.6721 - key_accuracy: 0.8710 - degree_1_accuracy: 0.9075 - degree_2_accuracy: 0.7216 - quality_accuracy: 0.7122 - inversion_accuracy: 0.5965 - root_accuracy: 0.7733 - val_loss: 4.5516 - val_key_loss: 0.7340 - val_degree_1_loss: 0.3628 - val_degree_2_loss: 0.9171 - val_quality_loss: 0.8768 - val_inversion_loss: 0.9288 - val_root_loss: 0.7321 - val_key_accuracy: 0.7698 - val_degree_1_accuracy: 0.9119 - val_degree_2_accuracy: 0.6785 - val_quality_accuracy: 0.6880 - val_inversion_accuracy: 0.6166 - val_root_accuracy: 0.7745

loss: 3.8639
key_loss: 0.3908
degree_1_loss: 0.3438
degree_2_loss: 0.7585
quality_loss: 0.7840
inversion_loss: 0.9146
root_loss: 0.6721

key_accuracy: 0.8710
degree_1_accuracy: 0.9075
degree_2_accuracy: 0.7216
quality_accuracy: 0.7122
inversion_accuracy: 0.5965
root_accuracy: 0.7733

val_loss: 4.5516
val_key_loss: 0.7340
val_degree_1_loss: 0.3628
val_degree_2_loss: 0.9171
val_quality_loss: 0.8768
val_inversion_loss: 0.9288
val_root_loss: 0.7321

val_key_accuracy: 0.7698
val_degree_1_accuracy: 0.9119
val_degree_2_accuracy: 0.6785
val_quality_accuracy: 0.6880
val_inversion_accuracy: 0.6166
val_root_accuracy: 0.7745

No real difference.

I am really really suspicious about the inversion. It seems to be 60% no matter what, but I would expect it to drop a lot once you lose information about the bass. It doesn't seem to be the case, and to me the most viable explanation is a bug in the code.

Today, I am learning abotu TFRecord representation they used to encode the dataset.

I want to feel comfortable "peeking" at the input data points that the model is reading as training examples. For two reasons:

It's a good practice to look at the data that the network is consuming right before it consumes it, as opposed to whenever it's being stored/pre-processed. Bugs may hide in that process and are untraceable unless you look at the data that the network is digesting right before it digests it. In this same line, I want to overfit the network to a "known" batch of examples, maybe even a single example. I really need to see the network getting the information it is supposed to get and overfit the inversion as it should if it's large enough.
I may want to implement an alternative (simple) model, just to see if that inversion accuracy is really that crappy or it is due to some bug in their model. This is still a challenge because the sequences of the output are 8th notes, the sequences of the input are 32th notes, so the data I read is probably not very straightforward applicable to a small model.

I am going to take a break from looking at the code of Micchi et al. for now.

Mostly to clear my mind.

Today, I will be looking at the corrected/latest BPS annotations curated by Mark Gotham.

Plans:

[x] Generate one voice-leading realization of each of these scores with general parameters (no unison, open position)
[x] Listen to every example to make sure they align with the original mxl scores available in Micchi et al. Just to make sure the annotations accurately align with the structure of the scores. Update: They don't align. Therefore, I generated my own csv tabular ground truth from the annotations of the artificial examples. Theoretically, the csvs from the score and from the artificial examples are incompatible, but, in their own dataset, they are correct. This triggered an unfruitful discussion. See below.
[x] Generate two sets of possible augmentations
- Rendering the RomanText with .show(). That is, no (or bad) voice leading
- Voice-leading realizations of the previous points
[x] Make a simple experiment training with these alternative training sets and see how well they do predicting on the validation

Running the experiment listed above.

Good voice leading. All the training examples of the BPS dataset. All the csv ground truth files I generated myself. The tools provided with the repo of Micchi et al. crashed when I tried to generate the csv files with them. Thus I extracted the lyrics from the annotated examples and generated the csv with those. It wasn't difficult.

Training on voice-lead artificial training set, validating on real BPS validation

Input representation was pitch_bass_cut because, at the same time, I want to know if pitch_hybrid_cut does better. Probably not, because it'd take into account voice leading, and as we know, block chord texture is far from what you usually find on a piano sonata. I expect Chromagram-like (pitch_bass_cut) to do better.

python train.py --input pitch_bass_cut --model gru

Epoch 37/100
 1/10 [==>...........................] - ETA: 4s - loss: 3.6810 - key_loss: 0.6983 - degree_1_loss: 0.2808 - degree_2_loss: 0.7660 - quality_loss: 0.7417 - inversion_loss: 0.7831 - root_loss: 0.4112 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 3.8825 - key_loss: 0.7596 - degree_1_loss: 0.3135 - degree_2_loss: 0.7793 - quality_loss: 0.7761 - inversion_loss: 0.8072 - root_loss: 0.4467 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 3.9257 - key_loss: 0.7675 - degree_1_loss: 0.3225 - degree_2_loss: 0.7869 - quality_loss: 0.7834 - inversion_loss: 0.8128 - root_loss: 0.4526 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 3.9553 - key_loss: 0.7772 - degree_1_loss: 0.3252 - degree_2_loss: 0.7883 - quality_loss: 0.7855 - inversion_loss: 0.8181 - root_loss: 0.4611 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 3.9586 - key_loss: 0.7793 - degree_1_loss: 0.3247 - degree_2_loss: 0.7869 - quality_loss: 0.7849 - inversion_loss: 0.8198 - root_loss: 0.4629 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 3.9596 - key_loss: 0.7763 - degree_1_loss: 0.3253 - degree_2_loss: 0.7874 - quality_loss: 0.7858 - inversion_loss: 0.8198 - root_loss: 0.4650 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 3.9630 - key_loss: 0.7756 - degree_1_loss: 0.3241 - degree_2_loss: 0.7878 - quality_loss: 0.7866 - inversion_loss: 0.8217 - root_loss: 0.4672 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 3.9740 - key_loss: 0.7785 - degree_1_loss: 0.3245 - degree_2_loss: 0.7892 - quality_loss: 0.7880 - inversion_loss: 0.8244 - root_loss: 0.4694 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 3.9878 - key_loss: 0.7821 - degree_1_loss: 0.3256 - degree_2_loss: 0.7920 - quality_loss: 0.7902 - inversion_loss: 0.8255 - root_loss: 0.4723 - key_accura10/10 [==============================] - ETA: 0s - loss: 3.9985 - key_loss: 0.7844 - degree_1_loss: 0.3267 - degree_2_loss: 0.7940 - quality_loss: 0.7916 - inversion_loss: 0.8270 - root_loss: 0.4748 - key_accura10/10 [==============================] - 6s 562ms/step - loss: 4.0072 - key_loss: 0.7862 - degree_1_loss: 0.3276 - degree_2_loss: 0.7957 - quality_loss: 0.7927 - inversion_loss: 0.8282 - root_loss: 0.4769 - key_accuracy: 0.7821 - degree_1_accuracy: 0.9171 - degree_2_accuracy: 0.7410 - quality_accuracy: 0.7370 - inversion_accuracy: 0.6079 - root_accuracy: 0.8722 - val_loss: 8.2571 - val_key_loss: 1.3593 - val_degree_1_loss: 0.4113 - val_degree_2_loss: 1.2396 - val_quality_loss: 1.2509 - val_inversion_loss: 3.0363 - val_root_loss: 0.9598 - val_key_accuracy: 0.5762 - val_degree_1_accuracy: 0.9097 - val_degree_2_accuracy: 0.5373 - val_quality_accuracy: 0.5188 - val_inversion_accuracy: 0.1994 - val_root_accuracy: 0.7008

loss: 4.0072
key_loss: 0.7862
degree_1_loss: 0.3276
degree_2_loss: 0.7957
quality_loss: 0.7927
inversion_loss: 0.8282
root_loss: 0.4769

key_accuracy: 0.7821
degree_1_accuracy: 0.9171
degree_2_accuracy: 0.7410
quality_accuracy: 0.7370
inversion_accuracy: 0.6079
root_accuracy: 0.8722

val_loss: 8.2571
val_key_loss: 1.3593
val_degree_1_loss: 0.4113
val_degree_2_loss: 1.2396
val_quality_loss: 1.2509
val_inversion_loss: 3.0363
val_root_loss: 0.9598

val_key_accuracy: 0.5762
val_degree_1_accuracy: 0.9097
val_degree_2_accuracy: 0.5373
val_quality_accuracy: 0.5188
val_inversion_accuracy: 0.1994
val_root_accuracy: 0.7008

Now pitch_hybrid_cut. In this one, N_OCTAVES is set to 5, centered on middle C.

Update: I re-ran this again and obtained 0.23 on the key accuracy, just as in the other training sets (plain and original_bps). I triple checked that the virtual environment was on, the same parameters were used, and everything was recomputed three times. 0.23 every time. How did I get 0.50 originally? Also, the number of epochs is dropping every time I run, first, it reached 39, now it is stopping after 16. There's something about and Machine Learning, no matter how hard I try, we just don't seem to come along at all...

python train.py --input pitch_hybrid_cut --model gru

Epoch 39/100
 1/10 [==>...........................] - ETA: 4s - loss: 4.4439 - key_loss: 1.0217 - degree_1_loss: 0.4014 - degree_2_loss: 0.8958 - quality_loss: 0.8459 - inversion_loss: 0.7525 - root_loss: 0.5266 - key_accura 2/10 [=====>........................] - ETA: 3s - loss: 4.3801 - key_loss: 0.9660 - degree_1_loss: 0.3930 - degree_2_loss: 0.8892 - quality_loss: 0.8422 - inversion_loss: 0.7818 - root_loss: 0.5080 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 4.3145 - key_loss: 0.9364 - degree_1_loss: 0.3809 - degree_2_loss: 0.8774 - quality_loss: 0.8428 - inversion_loss: 0.7889 - root_loss: 0.4882 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 4.2771 - key_loss: 0.9219 - degree_1_loss: 0.3705 - degree_2_loss: 0.8680 - quality_loss: 0.8439 - inversion_loss: 0.7988 - root_loss: 0.4740 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 4.2373 - key_loss: 0.9085 - degree_1_loss: 0.3642 - degree_2_loss: 0.8574 - quality_loss: 0.8393 - inversion_loss: 0.8033 - root_loss: 0.4646 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 4.1926 - key_loss: 0.8959 - degree_1_loss: 0.3576 - degree_2_loss: 0.8466 - quality_loss: 0.8330 - inversion_loss: 0.8053 - root_loss: 0.4542 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 4.1452 - key_loss: 0.8801 - degree_1_loss: 0.3536 - degree_2_loss: 0.8357 - quality_loss: 0.8244 - inversion_loss: 0.8073 - root_loss: 0.4440 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 4.1135 - key_loss: 0.8663 - degree_1_loss: 0.3510 - degree_2_loss: 0.8290 - quality_loss: 0.8208 - inversion_loss: 0.8091 - root_loss: 0.4373 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 4.0838 - key_loss: 0.8528 - degree_1_loss: 0.3491 - degree_2_loss: 0.8231 - quality_loss: 0.8168 - inversion_loss: 0.8102 - root_loss: 0.4318 - key_accura10/10 [==============================] - ETA: 0s - loss: 4.0654 - key_loss: 0.8439 - degree_1_loss: 0.3486 - degree_2_loss: 0.8193 - quality_loss: 0.8146 - inversion_loss: 0.8111 - root_loss: 0.4279 - key_accura10/10 [==============================] - 6s 557ms/step - loss: 4.0503 - key_loss: 0.8366 - degree_1_loss: 0.3482 - degree_2_loss: 0.8162 - quality_loss: 0.8129 - inversion_loss: 0.8119 - root_loss: 0.4247 - key_accuracy: 0.7772 - degree_1_accuracy: 0.9125 - degree_2_accuracy: 0.7382 - quality_accuracy: 0.7394 - inversion_accuracy: 0.6430 - root_accuracy: 0.8841 - val_loss: 8.3058 - val_key_loss: 1.4871 - val_degree_1_loss: 0.5194 - val_degree_2_loss: 1.3116 - val_quality_loss: 1.3113 - val_inversion_loss: 2.6592 - val_root_loss: 1.0174 - val_key_accuracy: 0.5042 - val_degree_1_accuracy: 0.8910 - val_degree_2_accuracy: 0.5155 - val_quality_accuracy: 0.4863 - val_inversion_accuracy: 0.2054 - val_root_accuracy: 0.6756

loss: 4.0503
key_loss: 0.8366
degree_1_loss: 0.3482
degree_2_loss: 0.8162
quality_loss: 0.8129
inversion_loss: 0.8119
root_loss: 0.4247

key_accuracy: 0.7772
degree_1_accuracy: 0.9125
degree_2_accuracy: 0.7382
quality_accuracy: 0.7394
inversion_accuracy: 0.6430
root_accuracy: 0.8841

val_loss: 8.3058
val_key_loss: 1.4871
val_degree_1_loss: 0.5194
val_degree_2_loss: 1.3116
val_quality_loss: 1.3113
val_inversion_loss: 2.6592
val_root_loss: 1.0174

val_key_accuracy: 0.5042
val_degree_1_accuracy: 0.8910
val_degree_2_accuracy: 0.5155
val_quality_accuracy: 0.4863
val_inversion_accuracy: 0.2054
val_root_accuracy: 0.6756

The same thing as before, except that N_OCTAVES is set to 1 (only middle C). I expect this to do a bit better on key and other features, and worse in inversion.

Epoch 33/100
 1/10 [==>...........................] - ETA: 4s - loss: 4.3125 - key_loss: 1.0340 - degree_1_loss: 0.2646 - degree_2_loss: 0.8433 - quality_loss: 0.8268 - inversion_loss: 0.9130 - root_loss: 0.4307 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 4.3603 - key_loss: 1.0292 - degree_1_loss: 0.2775 - degree_2_loss: 0.8749 - quality_loss: 0.8353 - inversion_loss: 0.8839 - root_loss: 0.4594 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 4.3798 - key_loss: 1.0007 - degree_1_loss: 0.2973 - degree_2_loss: 0.8989 - quality_loss: 0.8379 - inversion_loss: 0.8799 - root_loss: 0.4651 - key_accura 4/10 [===========>..................] - ETA: 2s - loss: 4.4223 - key_loss: 0.9840 - degree_1_loss: 0.3097 - degree_2_loss: 0.9200 - quality_loss: 0.8487 - inversion_loss: 0.8804 - root_loss: 0.4794 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 4.4197 - key_loss: 0.9734 - degree_1_loss: 0.3163 - degree_2_loss: 0.9249 - quality_loss: 0.8478 - inversion_loss: 0.8724 - root_loss: 0.4849 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 4.4170 - key_loss: 0.9677 - degree_1_loss: 0.3211 - degree_2_loss: 0.9270 - quality_loss: 0.8466 - inversion_loss: 0.8669 - root_loss: 0.4876 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 4.4168 - key_loss: 0.9649 - degree_1_loss: 0.3249 - degree_2_loss: 0.9271 - quality_loss: 0.8459 - inversion_loss: 0.8653 - root_loss: 0.4887 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 4.4171 - key_loss: 0.9637 - degree_1_loss: 0.3278 - degree_2_loss: 0.9261 - quality_loss: 0.8447 - inversion_loss: 0.8668 - root_loss: 0.4879 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 4.4198 - key_loss: 0.9616 - degree_1_loss: 0.3314 - degree_2_loss: 0.9258 - quality_loss: 0.8445 - inversion_loss: 0.8692 - root_loss: 0.4872 - key_accura10/10 [==============================] - ETA: 0s - loss: 4.4212 - key_loss: 0.9605 - degree_1_loss: 0.3350 - degree_2_loss: 0.9250 - quality_loss: 0.8435 - inversion_loss: 0.8705 - root_loss: 0.4865 - key_accura10/10 [==============================] - 5s 550ms/step - loss: 4.4223 - key_loss: 0.9597 - degree_1_loss: 0.3380 - degree_2_loss: 0.9243 - quality_loss: 0.8427 - inversion_loss: 0.8716 - root_loss: 0.4860 - key_accuracy: 0.7385 - degree_1_accuracy: 0.9235 - degree_2_accuracy: 0.6992 - quality_accuracy: 0.7443 - inversion_accuracy: 0.5726 - root_accuracy: 0.8742 - val_loss: 8.2027 - val_key_loss: 1.4862 - val_degree_1_loss: 0.4316 - val_degree_2_loss: 1.2907 - val_quality_loss: 1.2493 - val_inversion_loss: 2.8096 - val_root_loss: 0.9353 - val_key_accuracy: 0.5046 - val_degree_1_accuracy: 0.9117 - val_degree_2_accuracy: 0.5250 - val_quality_accuracy: 0.5512 - val_inversion_accuracy: 0.1884 - val_root_accuracy: 0.7046

loss: 4.4223
key_loss: 0.9597
degree_1_loss: 0.3380
degree_2_loss: 0.9243
quality_loss: 0.8427
inversion_loss: 0.8716
root_loss: 0.4860

key_accuracy: 0.7385
degree_1_accuracy: 0.9235
degree_2_accuracy: 0.6992
quality_accuracy: 0.7443
inversion_accuracy: 0.5726
root_accuracy: 0.8742

val_loss: 8.2027
val_key_loss: 1.4862
val_degree_1_loss: 0.4316
val_degree_2_loss: 1.2907
val_quality_loss: 1.2493
val_inversion_loss: 2.8096
val_root_loss: 0.9353

val_key_accuracy: 0.5046
val_degree_1_accuracy: 0.9117
val_degree_2_accuracy: 0.5250
val_quality_accuracy: 0.5512
val_inversion_accuracy: 0.1884
val_root_accuracy: 0.7046

Much worse in inversion. At least that's finally consistent with my expectations. It did worse on key (which I didn't think would happen), but it did a bit better on degree_1.

Now the same experiments with chords generated by music21 rather than the voice-leading algorithm.

If there is any hope of this paper working at all, these must be lower than the previous ones. Regardless of input representation or architecture.

If not, then, voice leading is useless for machine learning.

Small caveat: When I ran preprocessing on this data I got 0 warnings, whereas the voice leading examples threw many warnings about entries without annotation. It is possible there is a bug in either this files (plain) or the others (voice leading).

python train.py --input pitch_bass_cut --model gru

Epoch 44/100
 1/10 [==>...........................] - ETA: 4s - loss: 4.3757 - key_loss: 0.8012 - degree_1_loss: 0.4981 - degree_2_loss: 0.7909 - quality_loss: 0.8062 - inversion_loss: 0.9856 - root_loss: 0.4938 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 4.2584 - key_loss: 0.7865 - degree_1_loss: 0.4752 - degree_2_loss: 0.7729 - quality_loss: 0.7990 - inversion_loss: 0.9262 - root_loss: 0.4985 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 4.1748 - key_loss: 0.7656 - degree_1_loss: 0.4586 - degree_2_loss: 0.7689 - quality_loss: 0.7932 - inversion_loss: 0.8958 - root_loss: 0.4928 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 4.1135 - key_loss: 0.7484 - degree_1_loss: 0.4449 - degree_2_loss: 0.7643 - quality_loss: 0.7847 - inversion_loss: 0.8821 - root_loss: 0.4890 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 4.0690 - key_loss: 0.7382 - degree_1_loss: 0.4343 - degree_2_loss: 0.7613 - quality_loss: 0.7771 - inversion_loss: 0.8734 - root_loss: 0.4846 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 4.0357 - key_loss: 0.7272 - degree_1_loss: 0.4248 - degree_2_loss: 0.7606 - quality_loss: 0.7741 - inversion_loss: 0.8691 - root_loss: 0.4798 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 4.0029 - key_loss: 0.7197 - degree_1_loss: 0.4170 - degree_2_loss: 0.7590 - quality_loss: 0.7696 - inversion_loss: 0.8633 - root_loss: 0.4744 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 3.9808 - key_loss: 0.7178 - degree_1_loss: 0.4101 - degree_2_loss: 0.7575 - quality_loss: 0.7656 - inversion_loss: 0.8614 - root_loss: 0.4685 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 3.9603 - key_loss: 0.7157 - degree_1_loss: 0.4043 - degree_2_loss: 0.7546 - quality_loss: 0.7620 - inversion_loss: 0.8604 - root_loss: 0.4633 - key_accura10/10 [==============================] - ETA: 0s - loss: 3.9432 - key_loss: 0.7146 - degree_1_loss: 0.3998 - degree_2_loss: 0.7522 - quality_loss: 0.7586 - inversion_loss: 0.8589 - root_loss: 0.4592 - key_accura10/10 [==============================] - 6s 561ms/step - loss: 3.9292 - key_loss: 0.7136 - degree_1_loss: 0.3961 - degree_2_loss: 0.7502 - quality_loss: 0.7558 - inversion_loss: 0.8576 - root_loss: 0.4559 - key_accuracy: 0.8050 - degree_1_accuracy: 0.8969 - degree_2_accuracy: 0.7533 - quality_accuracy: 0.7612 - inversion_accuracy: 0.6034 - root_accuracy: 0.8669 - val_loss: 7.8277 - val_key_loss: 1.2577 - val_degree_1_loss: 0.4129 - val_degree_2_loss: 1.1647 - val_quality_loss: 1.0994 - val_inversion_loss: 2.9019 - val_root_loss: 0.9910 - val_key_accuracy: 0.6169 - val_degree_1_accuracy: 0.9085 - val_degree_2_accuracy: 0.5622 - val_quality_accuracy: 0.5795 - val_inversion_accuracy: 0.1944 - val_root_accuracy: 0.6910

loss: 3.9292
key_loss: 0.7136
degree_1_loss: 0.3961
degree_2_loss: 0.7502
quality_loss: 0.7558
inversion_loss: 0.8576
root_loss: 0.4559

key_accuracy: 0.8050
degree_1_accuracy: 0.8969
degree_2_accuracy: 0.7533
quality_accuracy: 0.7612
inversion_accuracy: 0.6034
root_accuracy: 0.8669

val_loss: 7.8277
val_key_loss: 1.2577
val_degree_1_loss: 0.4129
val_degree_2_loss: 1.1647
val_quality_loss: 1.0994
val_inversion_loss: 2.9019
val_root_loss: 0.9910

val_key_accuracy: 0.6169
val_degree_1_accuracy: 0.9085
val_degree_2_accuracy: 0.5622
val_quality_accuracy: 0.5795
val_inversion_accuracy: 0.1944
val_root_accuracy: 0.6910

Now pitch_hybrid_cut. N_OCTAVES=1 (middle C).

python train.py --input pitch_hybrid_cut --model gru

Epoch 39/100
 1/10 [==>...........................] - ETA: 4s - loss: 3.4964 - key_loss: 0.7077 - degree_1_loss: 0.2520 - degree_2_loss: 0.7554 - quality_loss: 0.6642 - inversion_loss: 0.8325 - root_loss: 0.2845 - key_accura 2/10 [=====>........................] - ETA: 3s - loss: 3.6208 - key_loss: 0.7619 - degree_1_loss: 0.2746 - degree_2_loss: 0.7613 - quality_loss: 0.6521 - inversion_loss: 0.8644 - root_loss: 0.3064 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 3.7207 - key_loss: 0.7961 - degree_1_loss: 0.2885 - degree_2_loss: 0.7706 - quality_loss: 0.6699 - inversion_loss: 0.8601 - root_loss: 0.3355 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 3.8005 - key_loss: 0.8262 - degree_1_loss: 0.3054 - degree_2_loss: 0.7757 - quality_loss: 0.6811 - inversion_loss: 0.8576 - root_loss: 0.3545 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 3.8482 - key_loss: 0.8421 - degree_1_loss: 0.3176 - degree_2_loss: 0.7807 - quality_loss: 0.6897 - inversion_loss: 0.8517 - root_loss: 0.3662 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 3.8820 - key_loss: 0.8484 - degree_1_loss: 0.3278 - degree_2_loss: 0.7857 - quality_loss: 0.6961 - inversion_loss: 0.8502 - root_loss: 0.3739 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 3.8951 - key_loss: 0.8471 - degree_1_loss: 0.3335 - degree_2_loss: 0.7882 - quality_loss: 0.6995 - inversion_loss: 0.8492 - root_loss: 0.3776 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 3.9098 - key_loss: 0.8459 - degree_1_loss: 0.3394 - degree_2_loss: 0.7900 - quality_loss: 0.7021 - inversion_loss: 0.8504 - root_loss: 0.3821 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 3.9182 - key_loss: 0.8441 - degree_1_loss: 0.3434 - degree_2_loss: 0.7897 - quality_loss: 0.7034 - inversion_loss: 0.8527 - root_loss: 0.3849 - key_accura10/10 [==============================] - ETA: 0s - loss: 3.9214 - key_loss: 0.8414 - degree_1_loss: 0.3463 - degree_2_loss: 0.7886 - quality_loss: 0.7040 - inversion_loss: 0.8549 - root_loss: 0.3863 - key_accura10/10 [==============================] - 6s 563ms/step - loss: 3.9240 - key_loss: 0.8391 - degree_1_loss: 0.3487 - degree_2_loss: 0.7876 - quality_loss: 0.7044 - inversion_loss: 0.8567 - root_loss: 0.3875 - key_accuracy: 0.7498 - degree_1_accuracy: 0.9129 - degree_2_accuracy: 0.7424 - quality_accuracy: 0.7990 - inversion_accuracy: 0.5860 - root_accuracy: 0.8947 - val_loss: 7.8142 - val_key_loss: 1.3623 - val_degree_1_loss: 0.4151 - val_degree_2_loss: 1.2590 - val_quality_loss: 1.1853 - val_inversion_loss: 2.6609 - val_root_loss: 0.9317 - val_key_accuracy: 0.5910 - val_degree_1_accuracy: 0.9081 - val_degree_2_accuracy: 0.5382 - val_quality_accuracy: 0.5986 - val_inversion_accuracy: 0.2018 - val_root_accuracy: 0.7041

loss: 3.9240
key_loss: 0.8391
degree_1_loss: 0.3487
degree_2_loss: 0.7876
quality_loss: 0.7044
inversion_loss: 0.8567
root_loss: 0.3875

key_accuracy: 0.7498
degree_1_accuracy: 0.9129
degree_2_accuracy: 0.7424
quality_accuracy: 0.7990
inversion_accuracy: 0.5860
root_accuracy: 0.8947

val_loss: 7.8142
val_key_loss: 1.3623
val_degree_1_loss: 0.4151
val_degree_2_loss: 1.2590
val_quality_loss: 1.1853
val_inversion_loss: 2.6609
val_root_loss: 0.9317

val_key_accuracy: 0.5910
val_degree_1_accuracy: 0.9081
val_degree_2_accuracy: 0.5382
val_quality_accuracy: 0.5986
val_inversion_accuracy: 0.2018
val_root_accuracy: 0.7041

Same but N_OCTAVES=5 (centered around middle C).

python train.py --input pitch_hybrid_cut --model gru

Epoch 23/100
 1/10 [==>...........................] - ETA: 4s - loss: 4.8824 - key_loss: 1.0830 - degree_1_loss: 0.4167 - degree_2_loss: 0.9160 - quality_loss: 0.9875 - inversion_loss: 0.9190 - root_loss: 0.5602 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 4.8967 - key_loss: 1.0875 - degree_1_loss: 0.4183 - degree_2_loss: 0.9173 - quality_loss: 0.9786 - inversion_loss: 0.9218 - root_loss: 0.5732 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 4.9074 - key_loss: 1.0864 - degree_1_loss: 0.4273 - degree_2_loss: 0.9261 - quality_loss: 0.9796 - inversion_loss: 0.9124 - root_loss: 0.5756 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 4.9113 - key_loss: 1.0952 - degree_1_loss: 0.4259 - degree_2_loss: 0.9318 - quality_loss: 0.9791 - inversion_loss: 0.9099 - root_loss: 0.5694 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 4.8720 - key_loss: 1.0865 - degree_1_loss: 0.4219 - degree_2_loss: 0.9274 - quality_loss: 0.9733 - inversion_loss: 0.9019 - root_loss: 0.5609 - key_accura 6/10 [=================>............] - ETA: 1s - loss: 4.8579 - key_loss: 1.0851 - degree_1_loss: 0.4193 - degree_2_loss: 0.9253 - quality_loss: 0.9752 - inversion_loss: 0.8957 - root_loss: 0.5574 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 4.8609 - key_loss: 1.0880 - degree_1_loss: 0.4209 - degree_2_loss: 0.9251 - quality_loss: 0.9787 - inversion_loss: 0.8902 - root_loss: 0.5580 - key_accura 8/10 [=======================>......] - ETA: 0s - loss: 4.8639 - key_loss: 1.0910 - degree_1_loss: 0.4214 - degree_2_loss: 0.9242 - quality_loss: 0.9821 - inversion_loss: 0.8867 - root_loss: 0.5586 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 4.8650 - key_loss: 1.0924 - degree_1_loss: 0.4205 - degree_2_loss: 0.9240 - quality_loss: 0.9856 - inversion_loss: 0.8838 - root_loss: 0.5586 - key_accura10/10 [==============================] - ETA: 0s - loss: 4.8719 - key_loss: 1.0955 - degree_1_loss: 0.4197 - degree_2_loss: 0.9246 - quality_loss: 0.9893 - inversion_loss: 0.8819 - root_loss: 0.5610 - key_accura10/10 [==============================] - 5s 547ms/step - loss: 4.8776 - key_loss: 1.0980 - degree_1_loss: 0.4190 - degree_2_loss: 0.9251 - quality_loss: 0.9924 - inversion_loss: 0.8803 - root_loss: 0.5629 - key_accuracy: 0.6953 - degree_1_accuracy: 0.9031 - degree_2_accuracy: 0.6879 - quality_accuracy: 0.6467 - inversion_accuracy: 0.5784 - root_accuracy: 0.8590 - val_loss: 10.1462 - val_key_loss: 2.2668 - val_degree_1_loss: 0.5165 - val_degree_2_loss: 1.6891 - val_quality_loss: 1.6460 - val_inversion_loss: 2.4763 - val_root_loss: 1.5516 - val_key_accuracy: 0.2686 - val_degree_1_accuracy: 0.9045 - val_degree_2_accuracy: 0.3756 - val_quality_accuracy: 0.3333 - val_inversion_accuracy: 0.1944 - val_root_accuracy: 0.4618

loss: 4.8776
key_loss: 1.0980
degree_1_loss: 0.4190
degree_2_loss: 0.9251
quality_loss: 0.9924
inversion_loss: 0.8803
root_loss: 0.5629

key_accuracy: 0.6953
degree_1_accuracy: 0.9031
degree_2_accuracy: 0.6879
quality_accuracy: 0.6467
inversion_accuracy: 0.5784
root_accuracy: 0.8590

val_loss: 10.1462
val_key_loss: 2.2668
val_degree_1_loss: 0.5165
val_degree_2_loss: 1.6891
val_quality_loss: 1.6460
val_inversion_loss: 2.4763
val_root_loss: 1.5516

val_key_accuracy: 0.2686
val_degree_1_accuracy: 0.9045
val_degree_2_accuracy: 0.3756
val_quality_accuracy: 0.3333
val_inversion_accuracy: 0.1944
val_root_accuracy: 0.4618

Now, something unexpected. The last one (5 octaves) did much much worse than the same model/representation using the voice leading artificial examples.

For reference, let's use the original beethoven training dataset. Same number of files, theoretically, same number of annotations. Now the scores in the training set come from the same distribution as the validation set.

Testing on the same three combinations: pitch_bass_cut, pitch_hybrid_cut(5 octaves), pitch_hybrid_cut(1 octave

python train.py --input pitch_bass_cut --model gru

Epoch 37/100
 1/10 [==>...........................] - ETA: 4s - loss: 8.2685 - key_loss: 1.9749 - degree_1_loss: 0.3868 - degree_2_loss: 1.4431 - quality_loss: 1.4536 - inversion_loss: 1.0215 - root_loss: 1.9886 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 7.9710 - key_loss: 1.8819 - degree_1_loss: 0.3776 - degree_2_loss: 1.4052 - quality_loss: 1.4309 - inversion_loss: 0.9946 - root_loss: 1.8807 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 7.8075 - key_loss: 1.8353 - degree_1_loss: 0.3779 - degree_2_loss: 1.3858 - quality_loss: 1.4150 - inversion_loss: 0.9802 - root_loss: 1.8132 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 7.7001 - key_loss: 1.7977 - degree_1_loss: 0.3775 - degree_2_loss: 1.3722 - quality_loss: 1.4025 - inversion_loss: 0.9746 - root_loss: 1.7756 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 7.6092 - key_loss: 1.7681 - degree_1_loss: 0.3791 - degree_2_loss: 1.3552 - quality_loss: 1.3900 - inversion_loss: 0.9716 - root_loss: 1.7452 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 7.5539 - key_loss: 1.7496 - degree_1_loss: 0.3826 - degree_2_loss: 1.3446 - quality_loss: 1.3823 - inversion_loss: 0.9700 - root_loss: 1.7247 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 7.5138 - key_loss: 1.7352 - degree_1_loss: 0.3850 - degree_2_loss: 1.3375 - quality_loss: 1.3765 - inversion_loss: 0.9668 - root_loss: 1.7127 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 7.4778 - key_loss: 1.7206 - degree_1_loss: 0.3873 - degree_2_loss: 1.3317 - quality_loss: 1.3711 - inversion_loss: 0.9644 - root_loss: 1.7027 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 7.4536 - key_loss: 1.7108 - degree_1_loss: 0.3900 - degree_2_loss: 1.3276 - quality_loss: 1.3661 - inversion_loss: 0.9623 - root_loss: 1.6967 - key_accura10/10 [==============================] - ETA: 0s - loss: 7.4313 - key_loss: 1.7012 - degree_1_loss: 0.3921 - degree_2_loss: 1.3241 - quality_loss: 1.3614 - inversion_loss: 0.9608 - root_loss: 1.6916 - key_accura10/10 [==============================] - 6s 566ms/step - loss: 7.4130 - key_loss: 1.6933 - degree_1_loss: 0.3938 - degree_2_loss: 1.3212 - quality_loss: 1.3576 - inversion_loss: 0.9596 - root_loss: 1.6875 - key_accuracy: 0.5024 - degree_1_accuracy: 0.9117 - degree_2_accuracy: 0.4769 - quality_accuracy: 0.4529 - inversion_accuracy: 0.6001 - root_accuracy: 0.4560 - val_loss: 7.3944 - val_key_loss: 1.8260 - val_degree_1_loss: 0.4412 - val_degree_2_loss: 1.3449 - val_quality_loss: 1.3427 - val_inversion_loss: 0.9819 - val_root_loss: 1.4577 - val_key_accuracy: 0.4866 - val_degree_1_accuracy: 0.9079 - val_degree_2_accuracy: 0.4304 - val_quality_accuracy: 0.4183 - val_inversion_accuracy: 0.6094 - val_root_accuracy: 0.5531

loss: 7.4130
key_loss: 1.6933
degree_1_loss: 0.3938
degree_2_loss: 1.3212
quality_loss: 1.3576
inversion_loss: 0.9596
root_loss: 1.6875

key_accuracy: 0.5024
degree_1_accuracy: 0.9117
degree_2_accuracy: 0.4769
quality_accuracy: 0.4529
inversion_accuracy: 0.6001
root_accuracy: 0.4560

val_loss: 7.3944
val_key_loss: 1.8260
val_degree_1_loss: 0.4412
val_degree_2_loss: 1.3449
val_quality_loss: 1.3427
val_inversion_loss: 0.9819
val_root_loss: 1.4577

val_key_accuracy: 0.4866
val_degree_1_accuracy: 0.9079
val_degree_2_accuracy: 0.4304
val_quality_accuracy: 0.4183
val_inversion_accuracy: 0.6094
val_root_accuracy: 0.5531

N_OCTAVES=5

python train.py --input pitch_hybrid_cut --model gru

Epoch 26/100
 1/10 [==>...........................] - ETA: 5s - loss: 7.9888 - key_loss: 2.0754 - degree_1_loss: 0.3554 - degree_2_loss: 1.3197 - quality_loss: 1.4135 - inversion_loss: 0.9385 - root_loss: 1.8862 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 7.7527 - key_loss: 1.9485 - degree_1_loss: 0.3685 - degree_2_loss: 1.3064 - quality_loss: 1.3793 - inversion_loss: 0.9231 - root_loss: 1.8270 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 7.6534 - key_loss: 1.8947 - degree_1_loss: 0.3797 - degree_2_loss: 1.3014 - quality_loss: 1.3590 - inversion_loss: 0.9253 - root_loss: 1.7932 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 7.5999 - key_loss: 1.8755 - degree_1_loss: 0.3856 - degree_2_loss: 1.2913 - quality_loss: 1.3475 - inversion_loss: 0.9225 - root_loss: 1.7775 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 7.5927 - key_loss: 1.8728 - degree_1_loss: 0.3906 - degree_2_loss: 1.2885 - quality_loss: 1.3457 - inversion_loss: 0.9214 - root_loss: 1.7738 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 7.5787 - key_loss: 1.8696 - degree_1_loss: 0.3942 - degree_2_loss: 1.2834 - quality_loss: 1.3417 - inversion_loss: 0.9230 - root_loss: 1.7668 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 7.5755 - key_loss: 1.8690 - degree_1_loss: 0.3995 - degree_2_loss: 1.2802 - quality_loss: 1.3401 - inversion_loss: 0.9244 - root_loss: 1.7623 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 7.5702 - key_loss: 1.8659 - degree_1_loss: 0.4017 - degree_2_loss: 1.2792 - quality_loss: 1.3380 - inversion_loss: 0.9254 - root_loss: 1.7601 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 7.5618 - key_loss: 1.8605 - degree_1_loss: 0.4040 - degree_2_loss: 1.2792 - quality_loss: 1.3362 - inversion_loss: 0.9257 - root_loss: 1.7563 - key_accura10/10 [==============================] - ETA: 0s - loss: 7.5560 - key_loss: 1.8555 - degree_1_loss: 0.4059 - degree_2_loss: 1.2794 - quality_loss: 1.3349 - inversion_loss: 0.9264 - root_loss: 1.7539 - key_accura10/10 [==============================] - 6s 602ms/step - loss: 7.5512 - key_loss: 1.8514 - degree_1_loss: 0.4075 - degree_2_loss: 1.2795 - quality_loss: 1.3339 - inversion_loss: 0.9269 - root_loss: 1.7519 - key_accuracy: 0.4519 - degree_1_accuracy: 0.9088 - degree_2_accuracy: 0.5089 - quality_accuracy: 0.4651 - inversion_accuracy: 0.6019 - root_accuracy: 0.4348 - val_loss: 8.7452 - val_key_loss: 2.5273 - val_degree_1_loss: 0.4938 - val_degree_2_loss: 1.4430 - val_quality_loss: 1.4378 - val_inversion_loss: 1.0087 - val_root_loss: 1.8346 - val_key_accuracy: 0.2310 - val_degree_1_accuracy: 0.9072 - val_degree_2_accuracy: 0.3855 - val_quality_accuracy: 0.3672 - val_inversion_accuracy: 0.6066 - val_root_accuracy: 0.357

loss: 7.5512
key_loss: 1.8514
degree_1_loss: 0.4075
degree_2_loss: 1.2795
quality_loss: 1.3339
inversion_loss: 0.9269
root_loss: 1.7519

key_accuracy: 0.4519
degree_1_accuracy: 0.9088
degree_2_accuracy: 0.5089
quality_accuracy: 0.4651
inversion_accuracy: 0.6019
root_accuracy: 0.4348

val_loss: 8.7452
val_key_loss: 2.5273
val_degree_1_loss: 0.4938
val_degree_2_loss: 1.4430
val_quality_loss: 1.4378
val_inversion_loss: 1.0087
val_root_loss: 1.8346

val_key_accuracy: 0.2310
val_degree_1_accuracy: 0.9072
val_degree_2_accuracy: 0.3855
val_quality_accuracy: 0.3672
val_inversion_accuracy: 0.6066
val_root_accuracy: 0.3572

N_OCTAVES=1

python train.py --input pitch_hybrid_cut --model gru

Epoch 42/100
 1/10 [==>...........................] - ETA: 5s - loss: 8.2196 - key_loss: 2.0342 - degree_1_loss: 0.5486 - degree_2_loss: 1.3009 - quality_loss: 1.3925 - inversion_loss: 1.0262 - root_loss: 1.9172 - key_accura 2/10 [=====>........................] - ETA: 4s - loss: 7.9113 - key_loss: 1.9304 - degree_1_loss: 0.4918 - degree_2_loss: 1.2836 - quality_loss: 1.3643 - inversion_loss: 0.9748 - root_loss: 1.8663 - key_accura 3/10 [========>.....................] - ETA: 3s - loss: 7.7610 - key_loss: 1.8657 - degree_1_loss: 0.4673 - degree_2_loss: 1.2922 - quality_loss: 1.3559 - inversion_loss: 0.9650 - root_loss: 1.8149 - key_accura 4/10 [===========>..................] - ETA: 3s - loss: 7.6178 - key_loss: 1.7961 - degree_1_loss: 0.4501 - degree_2_loss: 1.2940 - quality_loss: 1.3462 - inversion_loss: 0.9613 - root_loss: 1.7702 - key_accura 5/10 [==============>...............] - ETA: 2s - loss: 7.5734 - key_loss: 1.7648 - degree_1_loss: 0.4420 - degree_2_loss: 1.2992 - quality_loss: 1.3488 - inversion_loss: 0.9632 - root_loss: 1.7554 - key_accura 6/10 [=================>............] - ETA: 2s - loss: 7.5386 - key_loss: 1.7421 - degree_1_loss: 0.4363 - degree_2_loss: 1.3026 - quality_loss: 1.3491 - inversion_loss: 0.9639 - root_loss: 1.7446 - key_accura 7/10 [====================>.........] - ETA: 1s - loss: 7.4989 - key_loss: 1.7191 - degree_1_loss: 0.4318 - degree_2_loss: 1.3040 - quality_loss: 1.3489 - inversion_loss: 0.9648 - root_loss: 1.7303 - key_accura 8/10 [=======================>......] - ETA: 1s - loss: 7.4725 - key_loss: 1.7017 - degree_1_loss: 0.4307 - degree_2_loss: 1.3049 - quality_loss: 1.3496 - inversion_loss: 0.9653 - root_loss: 1.7202 - key_accura 9/10 [==========================>...] - ETA: 0s - loss: 7.4387 - key_loss: 1.6844 - degree_1_loss: 0.4296 - degree_2_loss: 1.3036 - quality_loss: 1.3477 - inversion_loss: 0.9639 - root_loss: 1.7095 - key_accura10/10 [==============================] - ETA: 0s - loss: 7.4034 - key_loss: 1.6675 - degree_1_loss: 0.4281 - degree_2_loss: 1.3014 - quality_loss: 1.3447 - inversion_loss: 0.9631 - root_loss: 1.6986 - key_accura10/10 [==============================] - 6s 567ms/step - loss: 7.3745 - key_loss: 1.6537 - degree_1_loss: 0.4269 - degree_2_loss: 1.2997 - quality_loss: 1.3422 - inversion_loss: 0.9624 - root_loss: 1.6897 - key_accuracy: 0.4870 - degree_1_accuracy: 0.9028 - degree_2_accuracy: 0.4819 - quality_accuracy: 0.4545 - inversion_accuracy: 0.5862 - root_accuracy: 0.4461 - val_loss: 7.2782 - val_key_loss: 1.8010 - val_degree_1_loss: 0.4523 - val_degree_2_loss: 1.3329 - val_quality_loss: 1.3521 - val_inversion_loss: 0.9663 - val_root_loss: 1.3735 - val_key_accuracy: 0.4585 - val_degree_1_accuracy: 0.9079 - val_degree_2_accuracy: 0.4552 - val_quality_accuracy: 0.4008 - val_inversion_accuracy: 0.6178 - val_root_accuracy: 0.5713

loss: 7.3745
key_loss: 1.6537
degree_1_loss: 0.4269
degree_2_loss: 1.2997
quality_loss: 1.3422
inversion_loss: 0.9624
root_loss: 1.6897

key_accuracy: 0.4870
degree_1_accuracy: 0.9028
degree_2_accuracy: 0.4819
quality_accuracy: 0.4545
inversion_accuracy: 0.5862
root_accuracy: 0.4461

val_loss: 7.2782
val_key_loss: 1.8010
val_degree_1_loss: 0.4523
val_degree_2_loss: 1.3329
val_quality_loss: 1.3521
val_inversion_loss: 0.9663
val_root_loss: 1.3735

val_key_accuracy: 0.4585
val_degree_1_accuracy: 0.9079
val_degree_2_accuracy: 0.4552
val_quality_accuracy: 0.4008
val_inversion_accuracy: 0.6178
val_root_accuracy: 0.5713

At this point, I don't even trust any of these results anymore.

The purpose of all of them is to just give a very rough outline of whether the general ideas will work or not (augmentation, augmentation with voice leading vs without, augmentation + original data, etc.)

Running the original training set (without transposition) plus the augmented data (plain).

python train.py --input pitch_bass_cut --model gru

Epoch 30/100
 1/20 [>.............................] - ETA: 10s - loss: 5.7937 - key_loss: 1.0788 - degree_1_loss: 0.4403 - degree_2_loss: 1.0160 - quality_loss: 1.0206 - inversion_loss: 1.0512 - root_loss: 1.1868 - key_accur 2/20 [==>...........................] - ETA: 9s - loss: 5.6738 - key_loss: 1.0525 - degree_1_loss: 0.3964 - degree_2_loss: 1.0048 - quality_loss: 1.0140 - inversion_loss: 1.0550 - root_loss: 1.1511 - key_accura 3/20 [===>..........................] - ETA: 8s - loss: 5.8280 - key_loss: 1.1270 - degree_1_loss: 0.3895 - degree_2_loss: 1.0314 - quality_loss: 1.0320 - inversion_loss: 1.0694 - root_loss: 1.1788 - key_accura 4/20 [=====>........................] - ETA: 7s - loss: 5.9002 - key_loss: 1.1722 - degree_1_loss: 0.3831 - degree_2_loss: 1.0409 - quality_loss: 1.0382 - inversion_loss: 1.0775 - root_loss: 1.1884 - key_accura 5/20 [======>.......................] - ETA: 7s - loss: 5.9409 - key_loss: 1.1956 - degree_1_loss: 0.3832 - degree_2_loss: 1.0470 - quality_loss: 1.0387 - inversion_loss: 1.0828 - root_loss: 1.1936 - key_accura 6/20 [========>.....................] - ETA: 7s - loss: 5.9693 - key_loss: 1.2068 - degree_1_loss: 0.3852 - degree_2_loss: 1.0544 - quality_loss: 1.0403 - inversion_loss: 1.0853 - root_loss: 1.1972 - key_accura 7/20 [=========>....................] - ETA: 6s - loss: 6.0014 - key_loss: 1.2204 - degree_1_loss: 0.3896 - degree_2_loss: 1.0590 - quality_loss: 1.0421 - inversion_loss: 1.0881 - root_loss: 1.2022 - key_accura 8/20 [===========>..................] - ETA: 6s - loss: 6.0299 - key_loss: 1.2314 - degree_1_loss: 0.3922 - degree_2_loss: 1.0622 - quality_loss: 1.0444 - inversion_loss: 1.0922 - root_loss: 1.2074 - key_accura 9/20 [============>.................] - ETA: 5s - loss: 6.0519 - key_loss: 1.2405 - degree_1_loss: 0.3932 - degree_2_loss: 1.0645 - quality_loss: 1.0465 - inversion_loss: 1.0966 - root_loss: 1.2105 - key_accura10/20 [==============>...............] - ETA: 5s - loss: 6.0662 - key_loss: 1.2472 - degree_1_loss: 0.3933 - degree_2_loss: 1.0655 - quality_loss: 1.0476 - inversion_loss: 1.1005 - root_loss: 1.2122 - key_accura11/20 [===============>..............] - ETA: 4s - loss: 6.0750 - key_loss: 1.2518 - degree_1_loss: 0.3928 - degree_2_loss: 1.0662 - quality_loss: 1.0478 - inversion_loss: 1.1033 - root_loss: 1.2131 - key_accura12/20 [=================>............] - ETA: 4s - loss: 6.0852 - key_loss: 1.2553 - degree_1_loss: 0.3929 - degree_2_loss: 1.0667 - quality_loss: 1.0488 - inversion_loss: 1.1058 - root_loss: 1.2156 - key_accura13/20 [==================>...........] - ETA: 3s - loss: 6.0889 - key_loss: 1.2575 - degree_1_loss: 0.3931 - degree_2_loss: 1.0663 - quality_loss: 1.0489 - inversion_loss: 1.1075 - root_loss: 1.2156 - key_accura14/20 [====================>.........] - ETA: 3s - loss: 6.0954 - key_loss: 1.2600 - degree_1_loss: 0.3934 - degree_2_loss: 1.0665 - quality_loss: 1.0496 - inversion_loss: 1.1095 - root_loss: 1.2163 - key_accura15/20 [=====================>........] - ETA: 2s - loss: 6.1014 - key_loss: 1.2623 - degree_1_loss: 0.3934 - degree_2_loss: 1.0665 - quality_loss: 1.0502 - inversion_loss: 1.1113 - root_loss: 1.2176 - key_accura16/20 [=======================>......] - ETA: 2s - loss: 6.1030 - key_loss: 1.2627 - degree_1_loss: 0.3935 - degree_2_loss: 1.0660 - quality_loss: 1.0505 - inversion_loss: 1.1128 - root_loss: 1.2176 - key_accura17/20 [========================>.....] - ETA: 1s - loss: 6.1081 - key_loss: 1.2642 - degree_1_loss: 0.3936 - degree_2_loss: 1.0659 - quality_loss: 1.0514 - inversion_loss: 1.1143 - root_loss: 1.2187 - key_accura18/20 [==========================>...] - ETA: 1s - loss: 6.1120 - key_loss: 1.2661 - degree_1_loss: 0.3932 - degree_2_loss: 1.0656 - quality_loss: 1.0521 - inversion_loss: 1.1156 - root_loss: 1.2194 - key_accura19/20 [===========================>..] - ETA: 0s - loss: 6.1134 - key_loss: 1.2670 - degree_1_loss: 0.3930 - degree_2_loss: 1.0649 - quality_loss: 1.0523 - inversion_loss: 1.1167 - root_loss: 1.2196 - key_accura20/20 [==============================] - ETA: 0s - loss: 6.1131 - key_loss: 1.2672 - degree_1_loss: 0.3927 - degree_2_loss: 1.0641 - quality_loss: 1.0522 - inversion_loss: 1.1177 - root_loss: 1.2192 - key_accura20/20 [==============================] - 11s 528ms/step - loss: 6.1129 - key_loss: 1.2673 - degree_1_loss: 0.3925 - degree_2_loss: 1.0634 - quality_loss: 1.0521 - inversion_loss: 1.1186 - root_loss: 1.2189 - key_accuracy: 0.6230 - degree_1_accuracy: 0.9055 - degree_2_accuracy: 0.6189 - quality_accuracy: 0.6220 - inversion_accuracy: 0.4340 - root_accuracy: 0.6655 - val_loss: 6.4379 - val_key_loss: 1.4088 - val_degree_1_loss: 0.4077 - val_degree_2_loss: 1.2008 - val_quality_loss: 1.1952 - val_inversion_loss: 1.1471 - val_root_loss: 1.0783 - val_key_accuracy: 0.5610 - val_degree_1_accuracy: 0.9084 - val_degree_2_accuracy: 0.5367 - val_quality_accuracy: 0.5224 - val_inversion_accuracy: 0.3502 - val_root_accuracy: 0.6798

loss: 6.1129
key_loss: 1.2673
degree_1_loss: 0.3925
degree_2_loss: 1.0634
quality_loss: 1.0521
inversion_loss: 1.1186
root_loss: 1.2189

key_accuracy: 0.6230
degree_1_accuracy: 0.9055
degree_2_accuracy: 0.6189
quality_accuracy: 0.6220
inversion_accuracy: 0.4340
root_accuracy: 0.6655

val_loss: 6.4379
val_key_loss: 1.4088
val_degree_1_loss: 0.4077
val_degree_2_loss: 1.2008
val_quality_loss: 1.1952
val_inversion_loss: 1.1471
val_root_loss: 1.0783

val_key_accuracy: 0.5610
val_degree_1_accuracy: 0.9084
val_degree_2_accuracy: 0.5367
val_quality_accuracy: 0.5224
val_inversion_accuracy: 0.3502
val_root_accuracy: 0.6798

python train.py --input pitch_hybrid_cut --model gru

N_OCTAVES=1. I don't trust the results I obtain with N_OCTAVES=5 anymore.

Epoch 43/100
 1/20 [>.............................] - ETA: 9s - loss: 5.2192 - key_loss: 1.0580 - degree_1_loss: 0.3071 - degree_2_loss: 0.8825 - quality_loss: 0.9350 - inversion_loss: 1.0556 - root_loss: 0.9810 - key_accura 2/20 [==>...........................] - ETA: 9s - loss: 5.0797 - key_loss: 0.9927 - degree_1_loss: 0.3230 - degree_2_loss: 0.8742 - quality_loss: 0.9052 - inversion_loss: 1.0569 - root_loss: 0.9276 - key_accura 3/20 [===>..........................] - ETA: 8s - loss: 5.1395 - key_loss: 1.0079 - degree_1_loss: 0.3238 - degree_2_loss: 0.8863 - quality_loss: 0.9087 - inversion_loss: 1.0741 - root_loss: 0.9388 - key_accura 4/20 [=====>........................] - ETA: 8s - loss: 5.1493 - key_loss: 1.0039 - degree_1_loss: 0.3299 - degree_2_loss: 0.8856 - quality_loss: 0.9053 - inversion_loss: 1.0817 - root_loss: 0.9429 - key_accura 5/20 [======>.......................] - ETA: 7s - loss: 5.1336 - key_loss: 0.9941 - degree_1_loss: 0.3311 - degree_2_loss: 0.8829 - quality_loss: 0.8977 - inversion_loss: 1.0869 - root_loss: 0.9409 - key_accura 6/20 [========>.....................] - ETA: 7s - loss: 5.1417 - key_loss: 0.9888 - degree_1_loss: 0.3343 - degree_2_loss: 0.8846 - quality_loss: 0.8967 - inversion_loss: 1.0918 - root_loss: 0.9456 - key_accura 7/20 [=========>....................] - ETA: 6s - loss: 5.1711 - key_loss: 0.9869 - degree_1_loss: 0.3391 - degree_2_loss: 0.8912 - quality_loss: 0.8998 - inversion_loss: 1.0966 - root_loss: 0.9576 - key_accura 8/20 [===========>..................] - ETA: 6s - loss: 5.2336 - key_loss: 1.0011 - degree_1_loss: 0.3428 - degree_2_loss: 0.9027 - quality_loss: 0.9083 - inversion_loss: 1.1020 - root_loss: 0.9767 - key_accura 9/20 [============>.................] - ETA: 5s - loss: 5.2800 - key_loss: 1.0100 - degree_1_loss: 0.3454 - degree_2_loss: 0.9112 - quality_loss: 0.9155 - inversion_loss: 1.1067 - root_loss: 0.9912 - key_accura10/20 [==============>...............] - ETA: 5s - loss: 5.3268 - key_loss: 1.0202 - degree_1_loss: 0.3477 - degree_2_loss: 0.9195 - quality_loss: 0.9226 - inversion_loss: 1.1109 - root_loss: 1.0059 - key_accura11/20 [===============>..............] - ETA: 4s - loss: 5.3612 - key_loss: 1.0290 - degree_1_loss: 0.3496 - degree_2_loss: 0.9250 - quality_loss: 0.9272 - inversion_loss: 1.1136 - root_loss: 1.0168 - key_accura12/20 [=================>............] - ETA: 4s - loss: 5.3933 - key_loss: 1.0392 - degree_1_loss: 0.3515 - degree_2_loss: 0.9296 - quality_loss: 0.9310 - inversion_loss: 1.1155 - root_loss: 1.0265 - key_accura13/20 [==================>...........] - ETA: 3s - loss: 5.4188 - key_loss: 1.0463 - degree_1_loss: 0.3532 - degree_2_loss: 0.9329 - quality_loss: 0.9341 - inversion_loss: 1.1174 - root_loss: 1.0348 - key_accura14/20 [====================>.........] - ETA: 3s - loss: 5.4467 - key_loss: 1.0547 - degree_1_loss: 0.3553 - degree_2_loss: 0.9361 - quality_loss: 0.9375 - inversion_loss: 1.1194 - root_loss: 1.0438 - key_accura15/20 [=====================>........] - ETA: 2s - loss: 5.4650 - key_loss: 1.0609 - degree_1_loss: 0.3563 - degree_2_loss: 0.9380 - quality_loss: 0.9396 - inversion_loss: 1.1199 - root_loss: 1.0503 - key_accura16/20 [=======================>......] - ETA: 2s - loss: 5.4804 - key_loss: 1.0661 - degree_1_loss: 0.3574 - degree_2_loss: 0.9394 - quality_loss: 0.9413 - inversion_loss: 1.1207 - root_loss: 1.0556 - key_accura17/20 [========================>.....] - ETA: 1s - loss: 5.4894 - key_loss: 1.0696 - degree_1_loss: 0.3580 - degree_2_loss: 0.9398 - quality_loss: 0.9419 - inversion_loss: 1.1216 - root_loss: 1.0585 - key_accura18/20 [==========================>...] - ETA: 1s - loss: 5.4992 - key_loss: 1.0731 - degree_1_loss: 0.3586 - degree_2_loss: 0.9406 - quality_loss: 0.9429 - inversion_loss: 1.1225 - root_loss: 1.0616 - key_accura19/20 [===========================>..] - ETA: 0s - loss: 5.5080 - key_loss: 1.0763 - degree_1_loss: 0.3592 - degree_2_loss: 0.9411 - quality_loss: 0.9436 - inversion_loss: 1.1231 - root_loss: 1.0647 - key_accura20/20 [==============================] - ETA: 0s - loss: 5.5168 - key_loss: 1.0793 - degree_1_loss: 0.3598 - degree_2_loss: 0.9418 - quality_loss: 0.9445 - inversion_loss: 1.1235 - root_loss: 1.0678 - key_accura20/20 [==============================] - 11s 530ms/step - loss: 5.5248 - key_loss: 1.0821 - degree_1_loss: 0.3604 - degree_2_loss: 0.9425 - quality_loss: 0.9453 - inversion_loss: 1.1240 - root_loss: 1.0706 - key_accuracy: 0.6949 - degree_1_accuracy: 0.9083 - degree_2_accuracy: 0.6716 - quality_accuracy: 0.6843 - inversion_accuracy: 0.4322 - root_accuracy: 0.7179 - val_loss: 6.3258 - val_key_loss: 1.4049 - val_degree_1_loss: 0.3944 - val_degree_2_loss: 1.2010 - val_quality_loss: 1.1546 - val_inversion_loss: 1.1378 - val_root_loss: 1.0330 - val_key_accuracy: 0.5792 - val_degree_1_accuracy: 0.9087 - val_degree_2_accuracy: 0.5415 - val_quality_accuracy: 0.5734 - val_inversion_accuracy: 0.3196 - val_root_accuracy: 0.6913

loss: 5.5248
key_loss: 1.0821
degree_1_loss: 0.3604
degree_2_loss: 0.9425
quality_loss: 0.9453
inversion_loss: 1.1240
root_loss: 1.0706

key_accuracy: 0.6949
degree_1_accuracy: 0.9083
degree_2_accuracy: 0.6716
quality_accuracy: 0.6843
inversion_accuracy: 0.4322
root_accuracy: 0.7179

val_loss: 6.3258
val_key_loss: 1.4049
val_degree_1_loss: 0.3944
val_degree_2_loss: 1.2010
val_quality_loss: 1.1546
val_inversion_loss: 1.1378
val_root_loss: 1.0330

val_key_accuracy: 0.5792
val_degree_1_accuracy: 0.9087
val_degree_2_accuracy: 0.5415
val_quality_accuracy: 0.5734
val_inversion_accuracy: 0.3196
val_root_accuracy: 0.6913

Running the original training set (without transposition) plus the augmented data (voicelead).

At this point, I know that the voicelead has alignment issues due to repetition bars, while the plain examples don't. Therefore, a portion of this data is corrupt and these results must be taken with a grain of salt. I do these experiments anyway because I want to get a quick glance at whether they increase accuracy when combined with the original training set.

In the end, I think I need to re-do these experiments on my network of my own, in a very thorough, systematic way. I just want to know if there is any hope of success with these "quick" tests.

python train.py --input pitch_bass_cut --model gru

Epoch 36/100
 1/20 [>.............................] - ETA: 10s - loss: 6.0635 - key_loss: 1.0406 - degree_1_loss: 0.3693 - degree_2_loss: 1.1166 - quality_loss: 1.0417 - inversion_loss: 1.1732 - root_loss: 1.3222 - key_accur 2/20 [==>...........................] - ETA: 8s - loss: 6.0176 - key_loss: 1.0977 - degree_1_loss: 0.3541 - degree_2_loss: 1.0780 - quality_loss: 1.0507 - inversion_loss: 1.1529 - root_loss: 1.2841 - key_accura 3/20 [===>..........................] - ETA: 8s - loss: 6.0343 - key_loss: 1.1253 - degree_1_loss: 0.3504 - degree_2_loss: 1.0711 - quality_loss: 1.0611 - inversion_loss: 1.1484 - root_loss: 1.2780 - key_accura 4/20 [=====>........................] - ETA: 8s - loss: 5.9507 - key_loss: 1.1189 - degree_1_loss: 0.3486 - degree_2_loss: 1.0489 - quality_loss: 1.0511 - inversion_loss: 1.1470 - root_loss: 1.2363 - key_accura 5/20 [======>.......................] - ETA: 7s - loss: 5.9254 - key_loss: 1.1197 - degree_1_loss: 0.3529 - degree_2_loss: 1.0410 - quality_loss: 1.0469 - inversion_loss: 1.1450 - root_loss: 1.2200 - key_accura 6/20 [========>.....................] - ETA: 7s - loss: 5.9144 - key_loss: 1.1267 - degree_1_loss: 0.3536 - degree_2_loss: 1.0370 - quality_loss: 1.0456 - inversion_loss: 1.1403 - root_loss: 1.2113 - key_accura 7/20 [=========>....................] - ETA: 6s - loss: 5.9035 - key_loss: 1.1291 - degree_1_loss: 0.3557 - degree_2_loss: 1.0344 - quality_loss: 1.0432 - inversion_loss: 1.1377 - root_loss: 1.2034 - key_accura 8/20 [===========>..................] - ETA: 6s - loss: 5.8832 - key_loss: 1.1273 - degree_1_loss: 0.3576 - degree_2_loss: 1.0307 - quality_loss: 1.0389 - inversion_loss: 1.1347 - root_loss: 1.1941 - key_accura 9/20 [============>.................] - ETA: 5s - loss: 5.8634 - key_loss: 1.1239 - degree_1_loss: 0.3590 - degree_2_loss: 1.0275 - quality_loss: 1.0348 - inversion_loss: 1.1317 - root_loss: 1.1864 - key_accura10/20 [==============>...............] - ETA: 5s - loss: 5.8516 - key_loss: 1.1249 - degree_1_loss: 0.3599 - degree_2_loss: 1.0246 - quality_loss: 1.0322 - inversion_loss: 1.1293 - root_loss: 1.1808 - key_accura11/20 [===============>..............] - ETA: 4s - loss: 5.8454 - key_loss: 1.1267 - degree_1_loss: 0.3605 - degree_2_loss: 1.0232 - quality_loss: 1.0305 - inversion_loss: 1.1273 - root_loss: 1.1771 - key_accura12/20 [=================>............] - ETA: 4s - loss: 5.8405 - key_loss: 1.1282 - degree_1_loss: 0.3616 - degree_2_loss: 1.0222 - quality_loss: 1.0286 - inversion_loss: 1.1264 - root_loss: 1.1734 - key_accura13/20 [==================>...........] - ETA: 3s - loss: 5.8442 - key_loss: 1.1311 - degree_1_loss: 0.3625 - degree_2_loss: 1.0226 - quality_loss: 1.0286 - inversion_loss: 1.1256 - root_loss: 1.1738 - key_accura14/20 [====================>.........] - ETA: 3s - loss: 5.8460 - key_loss: 1.1331 - degree_1_loss: 0.3626 - degree_2_loss: 1.0226 - quality_loss: 1.0286 - inversion_loss: 1.1251 - root_loss: 1.1740 - key_accura15/20 [=====================>........] - ETA: 2s - loss: 5.8436 - key_loss: 1.1334 - degree_1_loss: 0.3623 - degree_2_loss: 1.0223 - quality_loss: 1.0283 - inversion_loss: 1.1242 - root_loss: 1.1730 - key_accura16/20 [=======================>......] - ETA: 2s - loss: 5.8459 - key_loss: 1.1356 - degree_1_loss: 0.3624 - degree_2_loss: 1.0224 - quality_loss: 1.0285 - inversion_loss: 1.1235 - root_loss: 1.1735 - key_accura17/20 [========================>.....] - ETA: 1s - loss: 5.8477 - key_loss: 1.1377 - degree_1_loss: 0.3624 - degree_2_loss: 1.0222 - quality_loss: 1.0285 - inversion_loss: 1.1231 - root_loss: 1.1739 - key_accura18/20 [==========================>...] - ETA: 1s - loss: 5.8488 - key_loss: 1.1391 - degree_1_loss: 0.3625 - degree_2_loss: 1.0218 - quality_loss: 1.0283 - inversion_loss: 1.1229 - root_loss: 1.1742 - key_accura19/20 [===========================>..] - ETA: 0s - loss: 5.8493 - key_loss: 1.1404 - degree_1_loss: 0.3627 - degree_2_loss: 1.0211 - quality_loss: 1.0279 - inversion_loss: 1.1228 - root_loss: 1.1743 - key_accura20/20 [==============================] - ETA: 0s - loss: 5.8499 - key_loss: 1.1417 - degree_1_loss: 0.3628 - degree_2_loss: 1.0205 - quality_loss: 1.0276 - inversion_loss: 1.1228 - root_loss: 1.1745 - key_accura20/20 [==============================] - 11s 530ms/step - loss: 5.8503 - key_loss: 1.1428 - degree_1_loss: 0.3629 - degree_2_loss: 1.0200 - quality_loss: 1.0273 - inversion_loss: 1.1227 - root_loss: 1.1747 - key_accuracy: 0.6803 - degree_1_accuracy: 0.9095 - degree_2_accuracy: 0.6507 - quality_accuracy: 0.6447 - inversion_accuracy: 0.4349 - root_accuracy: 0.6885 - val_loss: 6.2445 - val_key_loss: 1.2358 - val_degree_1_loss: 0.4075 - val_degree_2_loss: 1.1850 - val_quality_loss: 1.1853 - val_inversion_loss: 1.1665 - val_root_loss: 1.0644 - val_key_accuracy: 0.6298 - val_degree_1_accuracy: 0.9090 - val_degree_2_accuracy: 0.5604 - val_quality_accuracy: 0.5225 - val_inversion_accuracy: 0.3101 - val_root_accuracy: 0.6888

loss: 5.8503
key_loss: 1.1428
degree_1_loss: 0.3629
degree_2_loss: 1.0200
quality_loss: 1.0273
inversion_loss: 1.1227
root_loss: 1.1747

key_accuracy: 0.6803
degree_1_accuracy: 0.9095
degree_2_accuracy: 0.6507
quality_accuracy: 0.6447
inversion_accuracy: 0.4349
root_accuracy: 0.6885

val_loss: 6.2445
val_key_loss: 1.2358
val_degree_1_loss: 0.4075
val_degree_2_loss: 1.1850
val_quality_loss: 1.1853
val_inversion_loss: 1.1665
val_root_loss: 1.0644

val_key_accuracy: 0.6298
val_degree_1_accuracy: 0.9090
val_degree_2_accuracy: 0.5604
val_quality_accuracy: 0.5225
val_inversion_accuracy: 0.3101
val_root_accuracy: 0.6888

python train.py --input pitch_hybrid_cut --model gru

Epoch 38/100
 1/20 [>.............................] - ETA: 10s - loss: 5.7515 - key_loss: 1.2589 - degree_1_loss: 0.3624 - degree_2_loss: 0.9443 - quality_loss: 0.9760 - inversion_loss: 1.0563 - root_loss: 1.1536 - key_accur 2/20 [==>...........................] - ETA: 9s - loss: 5.7462 - key_loss: 1.2196 - degree_1_loss: 0.3796 - degree_2_loss: 0.9548 - quality_loss: 0.9954 - inversion_loss: 1.0551 - root_loss: 1.1418 - key_accura 3/20 [===>..........................] - ETA: 8s - loss: 5.7538 - key_loss: 1.2042 - degree_1_loss: 0.3776 - degree_2_loss: 0.9647 - quality_loss: 1.0028 - inversion_loss: 1.0616 - root_loss: 1.1429 - key_accura 4/20 [=====>........................] - ETA: 8s - loss: 5.7813 - key_loss: 1.2019 - degree_1_loss: 0.3773 - degree_2_loss: 0.9782 - quality_loss: 1.0100 - inversion_loss: 1.0658 - root_loss: 1.1482 - key_accura 5/20 [======>.......................] - ETA: 7s - loss: 5.7679 - key_loss: 1.1852 - degree_1_loss: 0.3791 - degree_2_loss: 0.9820 - quality_loss: 1.0090 - inversion_loss: 1.0744 - root_loss: 1.1382 - key_accura 6/20 [========>.....................] - ETA: 7s - loss: 5.7553 - key_loss: 1.1690 - degree_1_loss: 0.3779 - degree_2_loss: 0.9856 - quality_loss: 1.0099 - inversion_loss: 1.0795 - root_loss: 1.1334 - key_accura 7/20 [=========>....................] - ETA: 6s - loss: 5.7286 - key_loss: 1.1547 - degree_1_loss: 0.3762 - degree_2_loss: 0.9846 - quality_loss: 1.0064 - inversion_loss: 1.0839 - root_loss: 1.1227 - key_accura 8/20 [===========>..................] - ETA: 6s - loss: 5.6903 - key_loss: 1.1393 - degree_1_loss: 0.3747 - degree_2_loss: 0.9804 - quality_loss: 1.0002 - inversion_loss: 1.0871 - root_loss: 1.1086 - key_accura 9/20 [============>.................] - ETA: 5s - loss: 5.6521 - key_loss: 1.1235 - degree_1_loss: 0.3717 - degree_2_loss: 0.9759 - quality_loss: 0.9939 - inversion_loss: 1.0911 - root_loss: 1.0960 - key_accura10/20 [==============>...............] - ETA: 5s - loss: 5.6254 - key_loss: 1.1110 - degree_1_loss: 0.3700 - degree_2_loss: 0.9734 - quality_loss: 0.9892 - inversion_loss: 1.0941 - root_loss: 1.0877 - key_accura11/20 [===============>..............] - ETA: 4s - loss: 5.6122 - key_loss: 1.1054 - degree_1_loss: 0.3686 - degree_2_loss: 0.9720 - quality_loss: 0.9860 - inversion_loss: 1.0965 - root_loss: 1.0837 - key_accura12/20 [=================>............] - ETA: 4s - loss: 5.6071 - key_loss: 1.1025 - degree_1_loss: 0.3677 - degree_2_loss: 0.9716 - quality_loss: 0.9841 - inversion_loss: 1.0988 - root_loss: 1.0825 - key_accura13/20 [==================>...........] - ETA: 3s - loss: 5.6203 - key_loss: 1.1050 - degree_1_loss: 0.3671 - degree_2_loss: 0.9746 - quality_loss: 0.9858 - inversion_loss: 1.1010 - root_loss: 1.0869 - key_accura14/20 [====================>.........] - ETA: 3s - loss: 5.6378 - key_loss: 1.1088 - degree_1_loss: 0.3667 - degree_2_loss: 0.9778 - quality_loss: 0.9883 - inversion_loss: 1.1033 - root_loss: 1.0930 - key_accura15/20 [=====================>........] - ETA: 2s - loss: 5.6556 - key_loss: 1.1127 - degree_1_loss: 0.3664 - degree_2_loss: 0.9810 - quality_loss: 0.9904 - inversion_loss: 1.1055 - root_loss: 1.0995 - key_accura16/20 [=======================>......] - ETA: 2s - loss: 5.6673 - key_loss: 1.1153 - degree_1_loss: 0.3660 - degree_2_loss: 0.9834 - quality_loss: 0.9916 - inversion_loss: 1.1072 - root_loss: 1.1037 - key_accura17/20 [========================>.....] - ETA: 1s - loss: 5.6758 - key_loss: 1.1173 - degree_1_loss: 0.3656 - degree_2_loss: 0.9854 - quality_loss: 0.9926 - inversion_loss: 1.1086 - root_loss: 1.1064 - key_accura18/20 [==========================>...] - ETA: 1s - loss: 5.6830 - key_loss: 1.1189 - degree_1_loss: 0.3653 - degree_2_loss: 0.9871 - quality_loss: 0.9932 - inversion_loss: 1.1101 - root_loss: 1.1084 - key_accura19/20 [===========================>..] - ETA: 0s - loss: 5.6903 - key_loss: 1.1208 - degree_1_loss: 0.3654 - degree_2_loss: 0.9886 - quality_loss: 0.9938 - inversion_loss: 1.1113 - root_loss: 1.1103 - key_accura20/20 [==============================] - ETA: 0s - loss: 5.6986 - key_loss: 1.1232 - degree_1_loss: 0.3654 - degree_2_loss: 0.9902 - quality_loss: 0.9946 - inversion_loss: 1.1125 - root_loss: 1.1127 - key_accura20/20 [==============================] - 11s 534ms/step - loss: 5.7061 - key_loss: 1.1253 - degree_1_loss: 0.3654 - degree_2_loss: 0.9917 - quality_loss: 0.9953 - inversion_loss: 1.1136 - root_loss: 1.1148 - key_accuracy: 0.6723 - degree_1_accuracy: 0.9089 - degree_2_accuracy: 0.6415 - quality_accuracy: 0.6522 - inversion_accuracy: 0.4306 - root_accuracy: 0.6969 - val_loss: 6.2816 - val_key_loss: 1.2977 - val_degree_1_loss: 0.4000 - val_degree_2_loss: 1.1948 - val_quality_loss: 1.1588 - val_inversion_loss: 1.1943 - val_root_loss: 1.0360 - val_key_accuracy: 0.6018 - val_degree_1_accuracy: 0.9097 - val_degree_2_accuracy: 0.5485 - val_quality_accuracy: 0.5604 - val_inversion_accuracy: 0.2743 - val_root_accuracy: 0.6962

loss: 5.7061
key_loss: 1.1253
degree_1_loss: 0.3654
degree_2_loss: 0.9917
quality_loss: 0.9953
inversion_loss: 1.1136
root_loss: 1.1148

key_accuracy: 0.6723
degree_1_accuracy: 0.9089
degree_2_accuracy: 0.6415
quality_accuracy: 0.6522
inversion_accuracy: 0.4306
root_accuracy: 0.6969

val_loss: 6.2816
val_key_loss: 1.2977
val_degree_1_loss: 0.4000
val_degree_2_loss: 1.1948
val_quality_loss: 1.1588
val_inversion_loss: 1.1943
val_root_loss: 1.0360

val_key_accuracy: 0.6018
val_degree_1_accuracy: 0.9097
val_degree_2_accuracy: 0.5485
val_quality_accuracy: 0.5604
val_inversion_accuracy: 0.2743
val_root_accuracy: 0.6962

In general, except for the inversion class, everything seems to go up in accuracy with original+augmentation. Regardless of whether we talk about the voicelead augmentation, or the plain augmentation.

A new set of experiments on this one. Before I meet with Ich and Mark.

Things I've learned so far:

The best representation is spelling_bass_cut.

My own version of that (19 features instead of 35) performs similarly. Sometimes slightly better, sometimess slightly worse. The main advantage of mine is that it can encode beyond 2-sharps and 2-flats.
Based on point (1), it is meaningless to do good or bad voice leading. The network can't see the voice leading.
Augmentation (with bad voice leading) seems to work well for all features except the inversion.
We can focus on
- The automatic voice leading aspect of the research
- The ML improvments/exploration aspect of the research
- They are not quite connected anymore

The new experiments are just to check more in depth the effect of augmentation on micchi's model with the full "paper" parameters: --input spelling_bass_cut --model conv_gru

Here is Micchi's trained with

BPS original set training/validation set
spelling_bass_cut and conv_gru

Epoch 25/100
  1/105 [..............................] - ETA: 1:08 - loss: 3.8112 - key_loss: 0.4755 - degree_1_loss: 0.4201 - degree_2_loss: 0.7254 - quality_loss: 0.6843 - inversion_loss: 0.8382 - root_loss: 0.6677 - key_ac  2/105 [..............................] - ETA: 59s - loss: 3.6603 - key_loss: 0.4552 - degree_1_loss: 0.3868 - degree_2_loss: 0.7051 - quality_loss: 0.6664 - inversion_loss: 0.8224 - root_loss: 0.6244 - key_acc  3/105 [..............................] - ETA: 58s - loss: 3.6516 - key_loss: 0.4432 - degree_1_loss: 0.3755 - degree_2_loss: 0.7135 - quality_loss: 0.6713 - inversion_loss: 0.8259 - root_loss: 0.6221 - key_acc  4/105 [>.............................] - ETA: 58s - loss: 3.6242 - key_loss: 0.4346 - degree_1_loss: 0.3660 - degree_2_loss: 0.7158 - quality_loss: 0.6708 - inversion_loss: 0.8229 - root_loss: 0.6141 - key_acc  5/105 [>.............................] - ETA: 58s - loss: 3.6126 - key_loss: 0.4313 - degree_1_loss: 0.3601 - degree_2_loss: 0.7179 - quality_loss: 0.6720 - inversion_loss: 0.8217 - root_loss: 0.6096 - key_acc  6/105 [>.............................] - ETA: 58s - loss: 3.6238 - key_loss: 0.4309 - degree_1_loss: 0.3569 - degree_2_loss: 0.7223 - quality_loss: 0.6804 - inversion_loss: 0.8210 - root_loss: 0.6123 - key_acc  7/105 [=>............................] - ETA: 57s - loss: 3.6129 - key_loss: 0.4286 - degree_1_loss: 0.3510 - degree_2_loss: 0.7223 - quality_loss: 0.6821 - inversion_loss: 0.8182 - root_loss: 0.6107 - key_acc  8/105 [=>............................] - ETA: 57s - loss: 3.6118 - key_loss: 0.4288 - degree_1_loss: 0.3480 - degree_2_loss: 0.7233 - quality_loss: 0.6841 - inversion_loss: 0.8174 - root_loss: 0.6102 - key_acc  9/105 [=>............................] - ETA: 56s - loss: 3.6139 - key_loss: 0.4298 - degree_1_loss: 0.3454 - degree_2_loss: 0.7251 - quality_loss: 0.6861 - inversion_loss: 0.8165 - root_loss: 0.6110 - key_acc 10/105 [=>............................] - ETA: 55s - loss: 3.6144 - key_loss: 0.4313 - degree_1_loss: 0.3435 - degree_2_loss: 0.7262 - quality_loss: 0.6878 - inversion_loss: 0.8148 - root_loss: 0.6108 - key_acc 11/105 [==>...........................] - ETA: 55s - loss: 3.6139 - key_loss: 0.4323 - degree_1_loss: 0.3425 - degree_2_loss: 0.7273 - quality_loss: 0.6891 - inversion_loss: 0.8127 - root_loss: 0.6100 - key_acc 12/105 [==>...........................] - ETA: 54s - loss: 3.6112 - key_loss: 0.4323 - degree_1_loss: 0.3417 - degree_2_loss: 0.7278 - quality_loss: 0.6896 - inversion_loss: 0.8108 - root_loss: 0.6091 - key_acc 13/105 [==>...........................] - ETA: 54s - loss: 3.6078 - key_loss: 0.4316 - degree_1_loss: 0.3409 - degree_2_loss: 0.7281 - quality_loss: 0.6898 - inversion_loss: 0.8092 - root_loss: 0.6081 - key_acc 14/105 [===>..........................] - ETA: 53s - loss: 3.6058 - key_loss: 0.4316 - degree_1_loss: 0.3404 - degree_2_loss: 0.7284 - quality_loss: 0.6901 - inversion_loss: 0.8079 - root_loss: 0.6074 - key_acc 15/105 [===>..........................] - ETA: 53s - loss: 3.6017 - key_loss: 0.4312 - degree_1_loss: 0.3396 - degree_2_loss: 0.7281 - quality_loss: 0.6900 - inversion_loss: 0.8067 - root_loss: 0.6061 - key_acc 16/105 [===>..........................] - ETA: 52s - loss: 3.6000 - key_loss: 0.4313 - degree_1_loss: 0.3394 - degree_2_loss: 0.7282 - quality_loss: 0.6904 - inversion_loss: 0.8055 - root_loss: 0.6052 - key_acc 17/105 [===>..........................] - ETA: 51s - loss: 3.5967 - key_loss: 0.4309 - degree_1_loss: 0.3391 - degree_2_loss: 0.7280 - quality_loss: 0.6906 - inversion_loss: 0.8039 - root_loss: 0.6042 - key_acc 18/105 [====>.........................] - ETA: 51s - loss: 3.5911 - key_loss: 0.4299 - degree_1_loss: 0.3382 - degree_2_loss: 0.7276 - quality_loss: 0.6904 - inversion_loss: 0.8019 - root_loss: 0.6030 - key_acc 19/105 [====>.........................] - ETA: 50s - loss: 3.5865 - key_loss: 0.4292 - degree_1_loss: 0.3376 - degree_2_loss: 0.7273 - quality_loss: 0.6902 - inversion_loss: 0.8003 - root_loss: 0.6018 - key_acc 20/105 [====>.........................] - ETA: 49s - loss: 3.5832 - key_loss: 0.4289 - degree_1_loss: 0.3371 - degree_2_loss: 0.7272 - quality_loss: 0.6902 - inversion_loss: 0.7990 - root_loss: 0.6008 - key_acc 21/105 [=====>........................] - ETA: 49s - loss: 3.5790 - key_loss: 0.4283 - degree_1_loss: 0.3364 - degree_2_loss: 0.7269 - quality_loss: 0.6898 - inversion_loss: 0.7980 - root_loss: 0.5996 - key_acc 22/105 [=====>........................] - ETA: 48s - loss: 3.5762 - key_loss: 0.4280 - degree_1_loss: 0.3359 - degree_2_loss: 0.7267 - quality_loss: 0.6895 - inversion_loss: 0.7973 - root_loss: 0.5987 - key_acc 23/105 [=====>........................] - ETA: 47s - loss: 3.5736 - key_loss: 0.4279 - degree_1_loss: 0.3353 - degree_2_loss: 0.7265 - quality_loss: 0.6893 - inversion_loss: 0.7967 - root_loss: 0.5978 - key_acc 24/105 [=====>........................] - ETA: 47s - loss: 3.5699 - key_loss: 0.4278 - degree_1_loss: 0.3346 - degree_2_loss: 0.7261 - quality_loss: 0.6887 - inversion_loss: 0.7960 - root_loss: 0.5966 - key_acc 25/105 [======>.......................] - ETA: 46s - loss: 3.5660 - key_loss: 0.4275 - degree_1_loss: 0.3340 - degree_2_loss: 0.7257 - quality_loss: 0.6882 - inversion_loss: 0.7953 - root_loss: 0.5954 - key_acc 26/105 [======>.......................] - ETA: 46s - loss: 3.5631 - key_loss: 0.4272 - degree_1_loss: 0.3334 - degree_2_loss: 0.7254 - quality_loss: 0.6878 - inversion_loss: 0.7948 - root_loss: 0.5945 - key_acc 27/105 [======>.......................] - ETA: 45s - loss: 3.5595 - key_loss: 0.4268 - degree_1_loss: 0.3328 - degree_2_loss: 0.7249 - quality_loss: 0.6872 - inversion_loss: 0.7942 - root_loss: 0.5936 - key_acc 28/105 [=======>......................] - ETA: 44s - loss: 3.5559 - key_loss: 0.4264 - degree_1_loss: 0.3322 - degree_2_loss: 0.7245 - quality_loss: 0.6866 - inversion_loss: 0.7936 - root_loss: 0.5927 - key_acc 29/105 [=======>......................] - ETA: 44s - loss: 3.5533 - key_loss: 0.4263 - degree_1_loss: 0.3318 - degree_2_loss: 0.7241 - quality_loss: 0.6861 - inversion_loss: 0.7930 - root_loss: 0.5920 - key_acc 30/105 [=======>......................] - ETA: 43s - loss: 3.5502 - key_loss: 0.4260 - degree_1_loss: 0.3315 - degree_2_loss: 0.7236 - quality_loss: 0.6855 - inversion_loss: 0.7924 - root_loss: 0.5912 - key_acc 31/105 [=======>......................] - ETA: 43s - loss: 3.5471 - key_loss: 0.4258 - degree_1_loss: 0.3312 - degree_2_loss: 0.7232 - quality_loss: 0.6850 - inversion_loss: 0.7916 - root_loss: 0.5904 - key_acc 32/105 [========>.....................] - ETA: 42s - loss: 3.5443 - key_loss: 0.4255 - degree_1_loss: 0.3310 - degree_2_loss: 0.7227 - quality_loss: 0.6844 - inversion_loss: 0.7910 - root_loss: 0.5897 - key_acc 33/105 [========>.....................] - ETA: 41s - loss: 3.5413 - key_loss: 0.4253 - degree_1_loss: 0.3308 - degree_2_loss: 0.7221 - quality_loss: 0.6838 - inversion_loss: 0.7904 - root_loss: 0.5889 - key_acc 34/105 [========>.....................] - ETA: 41s - loss: 3.5383 - key_loss: 0.4252 - degree_1_loss: 0.3307 - degree_2_loss: 0.7215 - quality_loss: 0.6831 - inversion_loss: 0.7898 - root_loss: 0.5882 - key_acc 35/105 [=========>....................] - ETA: 40s - loss: 3.5355 - key_loss: 0.4250 - degree_1_loss: 0.3305 - degree_2_loss: 0.7208 - quality_loss: 0.6825 - inversion_loss: 0.7892 - root_loss: 0.5875 - key_acc 36/105 [=========>....................] - ETA: 40s - loss: 3.5326 - key_loss: 0.4247 - degree_1_loss: 0.3304 - degree_2_loss: 0.7202 - quality_loss: 0.6819 - inversion_loss: 0.7886 - root_loss: 0.5868 - key_acc 37/105 [=========>....................] - ETA: 39s - loss: 3.5298 - key_loss: 0.4244 - degree_1_loss: 0.3302 - degree_2_loss: 0.7196 - quality_loss: 0.6813 - inversion_loss: 0.7880 - root_loss: 0.5862 - key_acc 38/105 [=========>....................] - ETA: 38s - loss: 3.5270 - key_loss: 0.4241 - degree_1_loss: 0.3301 - degree_2_loss: 0.7190 - quality_loss: 0.6807 - inversion_loss: 0.7875 - root_loss: 0.5856 - key_acc 39/105 [==========>...................] - ETA: 38s - loss: 3.5243 - key_loss: 0.4238 - degree_1_loss: 0.3299 - degree_2_loss: 0.7185 - quality_loss: 0.6801 - inversion_loss: 0.7870 - root_loss: 0.5851 - key_acc 40/105 [==========>...................] - ETA: 37s - loss: 3.5221 - key_loss: 0.4236 - degree_1_loss: 0.3298 - degree_2_loss: 0.7180 - quality_loss: 0.6796 - inversion_loss: 0.7865 - root_loss: 0.5846 - key_acc 41/105 [==========>...................] - ETA: 37s - loss: 3.5199 - key_loss: 0.4234 - degree_1_loss: 0.3296 - degree_2_loss: 0.7176 - quality_loss: 0.6791 - inversion_loss: 0.7861 - root_loss: 0.5841 - key_acc 42/105 [===========>..................] - ETA: 36s - loss: 3.5179 - key_loss: 0.4232 - degree_1_loss: 0.3295 - degree_2_loss: 0.7172 - quality_loss: 0.6786 - inversion_loss: 0.7857 - root_loss: 0.5837 - key_acc 43/105 [===========>..................] - ETA: 35s - loss: 3.5162 - key_loss: 0.4230 - degree_1_loss: 0.3294 - degree_2_loss: 0.7168 - quality_loss: 0.6782 - inversion_loss: 0.7854 - root_loss: 0.5834 - key_acc 44/105 [===========>..................] - ETA: 35s - loss: 3.5145 - key_loss: 0.4228 - degree_1_loss: 0.3292 - degree_2_loss: 0.7165 - quality_loss: 0.6778 - inversion_loss: 0.7851 - root_loss: 0.5831 - key_acc 45/105 [===========>..................] - ETA: 34s - loss: 3.5127 - key_loss: 0.4225 - degree_1_loss: 0.3291 - degree_2_loss: 0.7161 - quality_loss: 0.6774 - inversion_loss: 0.7848 - root_loss: 0.5827 - key_acc 46/105 [============>.................] - ETA: 34s - loss: 3.5107 - key_loss: 0.4222 - degree_1_loss: 0.3289 - degree_2_loss: 0.7157 - quality_loss: 0.6770 - inversion_loss: 0.7845 - root_loss: 0.5824 - key_acc 47/105 [============>.................] - ETA: 33s - loss: 3.5090 - key_loss: 0.4220 - degree_1_loss: 0.3288 - degree_2_loss: 0.7154 - quality_loss: 0.6766 - inversion_loss: 0.7842 - root_loss: 0.5821 - key_acc 48/105 [============>.................] - ETA: 33s - loss: 3.5073 - key_loss: 0.4217 - degree_1_loss: 0.3287 - degree_2_loss: 0.7150 - quality_loss: 0.6761 - inversion_loss: 0.7840 - root_loss: 0.5818 - key_acc 49/105 [=============>................] - ETA: 32s - loss: 3.5059 - key_loss: 0.4215 - degree_1_loss: 0.3285 - degree_2_loss: 0.7147 - quality_loss: 0.6758 - inversion_loss: 0.7838 - root_loss: 0.5816 - key_acc 50/105 [=============>................] - ETA: 31s - loss: 3.5045 - key_loss: 0.4213 - degree_1_loss: 0.3283 - degree_2_loss: 0.7145 - quality_loss: 0.6755 - inversion_loss: 0.7837 - root_loss: 0.5813 - key_acc 51/105 [=============>................] - ETA: 31s - loss: 3.5030 - key_loss: 0.4210 - degree_1_loss: 0.3281 - degree_2_loss: 0.7142 - quality_loss: 0.6751 - inversion_loss: 0.7835 - root_loss: 0.5811 - key_acc 52/105 [=============>................] - ETA: 30s - loss: 3.5015 - key_loss: 0.4207 - degree_1_loss: 0.3279 - degree_2_loss: 0.7139 - quality_loss: 0.6748 - inversion_loss: 0.7834 - root_loss: 0.5808 - key_acc 53/105 [==============>...............] - ETA: 30s - loss: 3.5001 - key_loss: 0.4204 - degree_1_loss: 0.3277 - degree_2_loss: 0.7136 - quality_loss: 0.6745 - inversion_loss: 0.7833 - root_loss: 0.5806 - key_acc 54/105 [==============>...............] - ETA: 29s - loss: 3.4987 - key_loss: 0.4201 - degree_1_loss: 0.3276 - degree_2_loss: 0.7133 - quality_loss: 0.6742 - inversion_loss: 0.7832 - root_loss: 0.5804 - key_acc 55/105 [==============>...............] - ETA: 28s - loss: 3.4974 - key_loss: 0.4198 - degree_1_loss: 0.3274 - degree_2_loss: 0.7130 - quality_loss: 0.6739 - inversion_loss: 0.7831 - root_loss: 0.5802 - key_acc 56/105 [===============>..............] - ETA: 28s - loss: 3.4961 - key_loss: 0.4194 - degree_1_loss: 0.3273 - degree_2_loss: 0.7128 - quality_loss: 0.6736 - inversion_loss: 0.7830 - root_loss: 0.5799 - key_acc 57/105 [===============>..............] - ETA: 27s - loss: 3.4949 - key_loss: 0.4191 - degree_1_loss: 0.3272 - degree_2_loss: 0.7125 - quality_loss: 0.6734 - inversion_loss: 0.7830 - root_loss: 0.5798 - key_acc 58/105 [===============>..............] - ETA: 27s - loss: 3.4939 - key_loss: 0.4189 - degree_1_loss: 0.3270 - degree_2_loss: 0.7123 - quality_loss: 0.6731 - inversion_loss: 0.7830 - root_loss: 0.5796 - key_acc 59/105 [===============>..............] - ETA: 26s - loss: 3.4932 - key_loss: 0.4186 - degree_1_loss: 0.3269 - degree_2_loss: 0.7121 - quality_loss: 0.6730 - inversion_loss: 0.7830 - root_loss: 0.5795 - key_acc 60/105 [================>.............] - ETA: 26s - loss: 3.4925 - key_loss: 0.4184 - degree_1_loss: 0.3268 - degree_2_loss: 0.7120 - quality_loss: 0.6729 - inversion_loss: 0.7831 - root_loss: 0.5794 - key_acc 61/105 [================>.............] - ETA: 25s - loss: 3.4918 - key_loss: 0.4181 - degree_1_loss: 0.3267 - degree_2_loss: 0.7119 - quality_loss: 0.6727 - inversion_loss: 0.7831 - root_loss: 0.5792 - key_acc 62/105 [================>.............] - ETA: 24s - loss: 3.4910 - key_loss: 0.4178 - degree_1_loss: 0.3266 - degree_2_loss: 0.7118 - quality_loss: 0.6726 - inversion_loss: 0.7832 - root_loss: 0.5791 - key_acc 63/105 [=================>............] - ETA: 24s - loss: 3.4905 - key_loss: 0.4175 - degree_1_loss: 0.3265 - degree_2_loss: 0.7117 - quality_loss: 0.6725 - inversion_loss: 0.7832 - root_loss: 0.5790 - key_acc 64/105 [=================>............] - ETA: 23s - loss: 3.4899 - key_loss: 0.4172 - degree_1_loss: 0.3264 - degree_2_loss: 0.7116 - quality_loss: 0.6724 - inversion_loss: 0.7833 - root_loss: 0.5790 - key_acc 65/105 [=================>............] - ETA: 23s - loss: 3.4893 - key_loss: 0.4170 - degree_1_loss: 0.3263 - degree_2_loss: 0.7115 - quality_loss: 0.6723 - inversion_loss: 0.7834 - root_loss: 0.5789 - key_acc 66/105 [=================>............] - ETA: 22s - loss: 3.4887 - key_loss: 0.4167 - degree_1_loss: 0.3262 - degree_2_loss: 0.7114 - quality_loss: 0.6722 - inversion_loss: 0.7834 - root_loss: 0.5787 - key_acc 67/105 [==================>...........] - ETA: 21s - loss: 3.4882 - key_loss: 0.4165 - degree_1_loss: 0.3262 - degree_2_loss: 0.7113 - quality_loss: 0.6722 - inversion_loss: 0.7835 - root_loss: 0.5786 - key_acc 68/105 [==================>...........] - ETA: 21s - loss: 3.4877 - key_loss: 0.4162 - degree_1_loss: 0.3261 - degree_2_loss: 0.7112 - quality_loss: 0.6721 - inversion_loss: 0.7836 - root_loss: 0.5785 - key_acc 69/105 [==================>...........] - ETA: 20s - loss: 3.4871 - key_loss: 0.4159 - degree_1_loss: 0.3260 - degree_2_loss: 0.7110 - quality_loss: 0.6720 - inversion_loss: 0.7837 - root_loss: 0.5784 - key_acc 70/105 [===================>..........] - ETA: 20s - loss: 3.4866 - key_loss: 0.4157 - degree_1_loss: 0.3260 - degree_2_loss: 0.7109 - quality_loss: 0.6720 - inversion_loss: 0.7838 - root_loss: 0.5783 - key_acc 71/105 [===================>..........] - ETA: 19s - loss: 3.4861 - key_loss: 0.4154 - degree_1_loss: 0.3259 - degree_2_loss: 0.7108 - quality_loss: 0.6719 - inversion_loss: 0.7839 - root_loss: 0.5782 - key_acc 72/105 [===================>..........] - ETA: 19s - loss: 3.4855 - key_loss: 0.4152 - degree_1_loss: 0.3258 - degree_2_loss: 0.7106 - quality_loss: 0.6719 - inversion_loss: 0.7839 - root_loss: 0.5781 - key_acc 73/105 [===================>..........] - ETA: 18s - loss: 3.4849 - key_loss: 0.4149 - degree_1_loss: 0.3257 - degree_2_loss: 0.7104 - quality_loss: 0.6718 - inversion_loss: 0.7840 - root_loss: 0.5780 - key_acc 74/105 [====================>.........] - ETA: 17s - loss: 3.4842 - key_loss: 0.4147 - degree_1_loss: 0.3256 - degree_2_loss: 0.7103 - quality_loss: 0.6717 - inversion_loss: 0.7841 - root_loss: 0.5779 - key_acc 75/105 [====================>.........] - ETA: 17s - loss: 3.4836 - key_loss: 0.4144 - degree_1_loss: 0.3255 - degree_2_loss: 0.7101 - quality_loss: 0.6716 - inversion_loss: 0.7842 - root_loss: 0.5778 - key_acc 76/105 [====================>.........] - ETA: 16s - loss: 3.4829 - key_loss: 0.4141 - degree_1_loss: 0.3255 - degree_2_loss: 0.7099 - quality_loss: 0.6715 - inversion_loss: 0.7842 - root_loss: 0.5776 - key_acc 77/105 [=====================>........] - ETA: 16s - loss: 3.4824 - key_loss: 0.4139 - degree_1_loss: 0.3254 - degree_2_loss: 0.7098 - quality_loss: 0.6714 - inversion_loss: 0.7843 - root_loss: 0.5775 - key_acc 78/105 [=====================>........] - ETA: 15s - loss: 3.4818 - key_loss: 0.4137 - degree_1_loss: 0.3253 - degree_2_loss: 0.7096 - quality_loss: 0.6713 - inversion_loss: 0.7844 - root_loss: 0.5774 - key_acc 79/105 [=====================>........] - ETA: 14s - loss: 3.4813 - key_loss: 0.4134 - degree_1_loss: 0.3253 - degree_2_loss: 0.7094 - quality_loss: 0.6712 - inversion_loss: 0.7845 - root_loss: 0.5774 - key_acc 80/105 [=====================>........] - ETA: 14s - loss: 3.4808 - key_loss: 0.4132 - degree_1_loss: 0.3252 - degree_2_loss: 0.7093 - quality_loss: 0.6711 - inversion_loss: 0.7846 - root_loss: 0.5773 - key_acc 81/105 [======================>.......] - ETA: 13s - loss: 3.4803 - key_loss: 0.4130 - degree_1_loss: 0.3252 - degree_2_loss: 0.7091 - quality_loss: 0.6710 - inversion_loss: 0.7846 - root_loss: 0.5772 - key_acc 82/105 [======================>.......] - ETA: 13s - loss: 3.4798 - key_loss: 0.4129 - degree_1_loss: 0.3252 - degree_2_loss: 0.7090 - quality_loss: 0.6709 - inversion_loss: 0.7847 - root_loss: 0.5771 - key_acc 83/105 [======================>.......] - ETA: 12s - loss: 3.4794 - key_loss: 0.4127 - degree_1_loss: 0.3252 - degree_2_loss: 0.7088 - quality_loss: 0.6708 - inversion_loss: 0.7848 - root_loss: 0.5770 - key_acc 84/105 [=======================>......] - ETA: 12s - loss: 3.4789 - key_loss: 0.4125 - degree_1_loss: 0.3252 - degree_2_loss: 0.7087 - quality_loss: 0.6707 - inversion_loss: 0.7849 - root_loss: 0.5769 - key_acc 85/105 [=======================>......] - ETA: 11s - loss: 3.4784 - key_loss: 0.4122 - degree_1_loss: 0.3252 - degree_2_loss: 0.7085 - quality_loss: 0.6706 - inversion_loss: 0.7850 - root_loss: 0.5769 - key_acc 86/105 [=======================>......] - ETA: 10s - loss: 3.4780 - key_loss: 0.4121 - degree_1_loss: 0.3252 - degree_2_loss: 0.7084 - quality_loss: 0.6706 - inversion_loss: 0.7851 - root_loss: 0.5768 - key_acc 87/105 [=======================>......] - ETA: 10s - loss: 3.4775 - key_loss: 0.4118 - degree_1_loss: 0.3251 - degree_2_loss: 0.7082 - quality_loss: 0.6705 - inversion_loss: 0.7851 - root_loss: 0.5767 - key_acc 88/105 [========================>.....] - ETA: 9s - loss: 3.4771 - key_loss: 0.4116 - degree_1_loss: 0.3251 - degree_2_loss: 0.7081 - quality_loss: 0.6704 - inversion_loss: 0.7852 - root_loss: 0.5766 - key_accu 89/105 [========================>.....] - ETA: 9s - loss: 3.4767 - key_loss: 0.4114 - degree_1_loss: 0.3251 - degree_2_loss: 0.7080 - quality_loss: 0.6704 - inversion_loss: 0.7853 - root_loss: 0.5766 - key_accu 90/105 [========================>.....] - ETA: 8s - loss: 3.4763 - key_loss: 0.4112 - degree_1_loss: 0.3250 - degree_2_loss: 0.7078 - quality_loss: 0.6703 - inversion_loss: 0.7854 - root_loss: 0.5765 - key_accu 91/105 [=========================>....] - ETA: 8s - loss: 3.4758 - key_loss: 0.4110 - degree_1_loss: 0.3250 - degree_2_loss: 0.7077 - quality_loss: 0.6702 - inversion_loss: 0.7855 - root_loss: 0.5764 - key_accu 92/105 [=========================>....] - ETA: 7s - loss: 3.4754 - key_loss: 0.4108 - degree_1_loss: 0.3250 - degree_2_loss: 0.7075 - quality_loss: 0.6702 - inversion_loss: 0.7855 - root_loss: 0.5763 - key_accu 93/105 [=========================>....] - ETA: 6s - loss: 3.4748 - key_loss: 0.4106 - degree_1_loss: 0.3249 - degree_2_loss: 0.7074 - quality_loss: 0.6701 - inversion_loss: 0.7856 - root_loss: 0.5762 - key_accu 94/105 [=========================>....] - ETA: 6s - loss: 3.4742 - key_loss: 0.4104 - degree_1_loss: 0.3249 - degree_2_loss: 0.7072 - quality_loss: 0.6700 - inversion_loss: 0.7856 - root_loss: 0.5761 - key_accu 95/105 [==========================>...] - ETA: 5s - loss: 3.4737 - key_loss: 0.4102 - degree_1_loss: 0.3249 - degree_2_loss: 0.7070 - quality_loss: 0.6699 - inversion_loss: 0.7856 - root_loss: 0.5760 - key_accu 96/105 [==========================>...] - ETA: 5s - loss: 3.4732 - key_loss: 0.4100 - degree_1_loss: 0.3248 - degree_2_loss: 0.7069 - quality_loss: 0.6698 - inversion_loss: 0.7857 - root_loss: 0.5760 - key_accu 97/105 [==========================>...] - ETA: 4s - loss: 3.4727 - key_loss: 0.4099 - degree_1_loss: 0.3248 - degree_2_loss: 0.7067 - quality_loss: 0.6697 - inversion_loss: 0.7857 - root_loss: 0.5759 - key_accu 98/105 [===========================>..] - ETA: 4s - loss: 3.4722 - key_loss: 0.4097 - degree_1_loss: 0.3248 - degree_2_loss: 0.7066 - quality_loss: 0.6696 - inversion_loss: 0.7858 - root_loss: 0.5758 - key_accu 99/105 [===========================>..] - ETA: 3s - loss: 3.4718 - key_loss: 0.4095 - degree_1_loss: 0.3247 - degree_2_loss: 0.7064 - quality_loss: 0.6696 - inversion_loss: 0.7858 - root_loss: 0.5757 - key_accu100/105 [===========================>..] - ETA: 2s - loss: 3.4713 - key_loss: 0.4093 - degree_1_loss: 0.3247 - degree_2_loss: 0.7063 - quality_loss: 0.6695 - inversion_loss: 0.7859 - root_loss: 0.5757 - key_accu101/105 [===========================>..] - ETA: 2s - loss: 3.4708 - key_loss: 0.4091 - degree_1_loss: 0.3247 - degree_2_loss: 0.7061 - quality_loss: 0.6694 - inversion_loss: 0.7859 - root_loss: 0.5756 - key_accu102/105 [============================>.] - ETA: 1s - loss: 3.4703 - key_loss: 0.4089 - degree_1_loss: 0.3247 - degree_2_loss: 0.7059 - quality_loss: 0.6693 - inversion_loss: 0.7860 - root_loss: 0.5755 - key_accu103/105 [============================>.] - ETA: 1s - loss: 3.4699 - key_loss: 0.4088 - degree_1_loss: 0.3246 - degree_2_loss: 0.7058 - quality_loss: 0.6692 - inversion_loss: 0.7860 - root_loss: 0.5755 - key_accu104/105 [============================>.] - ETA: 0s - loss: 3.4693 - key_loss: 0.4086 - degree_1_loss: 0.3246 - degree_2_loss: 0.7056 - quality_loss: 0.6691 - inversion_loss: 0.7860 - root_loss: 0.5754 - key_accu105/105 [==============================] - ETA: 0s - loss: 3.4688 - key_loss: 0.4084 - degree_1_loss: 0.3246 - degree_2_loss: 0.7055 - quality_loss: 0.6690 - inversion_loss: 0.7861 - root_loss: 0.5753 - key_accu105/105 [==============================] - 61s 584ms/step - loss: 3.4683 - key_loss: 0.4082 - degree_1_loss: 0.3245 - degree_2_loss: 0.7053 - quality_loss: 0.6689 - inversion_loss: 0.7861 - root_loss: 0.5753 - key_accuracy: 0.8720 - degree_1_accuracy: 0.9095 - degree_2_accuracy: 0.7494 - quality_accuracy: 0.7665 - inversion_accuracy: 0.6694 - root_accuracy: 0.8134 - val_loss: 4.2601 - val_key_loss: 0.6988 - val_degree_1_loss: 0.3566 - val_degree_2_loss: 0.8909 - val_quality_loss: 0.7931 - val_inversion_loss: 0.8543 - val_root_loss: 0.6663 - val_key_accuracy: 0.7693 - val_degree_1_accuracy: 0.9101 - val_degree_2_accuracy: 0.6949 - val_quality_accuracy: 0.7491 - val_inversion_accuracy: 0.6507 - val_root_accuracy: 0.7994

loss: 3.4683
key_loss: 0.4082
degree_1_loss: 0.3245
degree_2_loss: 0.7053
quality_loss: 0.6689
inversion_loss: 0.7861
root_loss: 0.5753

key_accuracy: 0.8720
degree_1_accuracy: 0.9095
degree_2_accuracy: 0.7494
quality_accuracy: 0.7665
inversion_accuracy: 0.6694
root_accuracy: 0.8134

val_loss: 4.2601
val_key_loss: 0.6988
val_degree_1_loss: 0.3566
val_degree_2_loss: 0.8909
val_quality_loss: 0.7931
val_inversion_loss: 0.8543
val_root_loss: 0.6663

val_key_accuracy: 0.7693
val_degree_1_accuracy: 0.9101
val_degree_2_accuracy: 0.6949
val_quality_accuracy: 0.7491
val_inversion_accuracy: 0.6507
val_root_accuracy: 0.7994

Here is what the output of analyse_results looks like:

results = {
    'key': 76.92756630568705
    'degree 1': 91.01225601452565
    'degree 2': 69.48965696128656
    'quality': 74.91083587315998
    'inversion': 65.0671162700214
    'root': 79.94293495882239
    'degree': 63.322741715842035
    'secondary': 7.648725212464589
    'derived root': 67.44698787367875
    'roman': 49.82815641009014
    'roman + inv': 33.156085856948316
    'root coherence': 74.3920627715453
    'd7 no inv': 17.885117493472585
}

Buggy version of experiment "BPSSynth on BPSSynth-val", before fixing issue with `expandRepeats()`

It was laborious, but I am confident that I am generating a robust `BPSSynth` dataset now. It is made of the RomanText "originals" of BPS, but curated/sanitized by Mark Gotham and the `mxl` they generated when parsed. Micchi's code is still doing weird things with the alignment, but I am confident that this is on his end and the scores-annotations align perfectly in the dataset. Here is running Micchi's model on BPSSynth-val, with regular parameters: - BPSSynth original set training/validation split. All scores are synthetic (even validation). - `spelling_bass_cut` and `conv_gru` ```python Epoch 11/100 1/93 [..............................] - ETA: 1:00 - loss: 4.3315 - key_loss: 0.9063 - degree_1_loss: 0.2157 - degree_2_loss: 0.8375 - quality_loss: 0.7851 - inversion_loss: 0.5862 - root_loss: 1.0007 - key_accu 2/93 [..............................] - ETA: 52s - loss: 4.4409 - key_loss: 0.9369 - degree_1_loss: 0.2460 - degree_2_loss: 0.8645 - quality_loss: 0.7843 - inversion_loss: 0.6026 - root_loss: 1.0065 - key_accur 3/93 [..............................] - ETA: 53s - loss: 4.5015 - key_loss: 0.9379 - degree_1_loss: 0.2639 - degree_2_loss: 0.8842 - quality_loss: 0.7877 - inversion_loss: 0.6332 - root_loss: 0.9948 - key_accur 4/93 [>.............................] - ETA: 52s - loss: 4.5439 - key_loss: 0.9448 - degree_1_loss: 0.2791 - degree_2_loss: 0.8936 - quality_loss: 0.7860 - inversion_loss: 0.6549 - root_loss: 0.9855 - key_accur 5/93 [>.............................] - ETA: 51s - loss: 4.5234 - key_loss: 0.9358 - degree_1_loss: 0.2825 - degree_2_loss: 0.8928 - quality_loss: 0.7808 - inversion_loss: 0.6629 - root_loss: 0.9685 - key_accur 6/93 [>.............................] - ETA: 51s - loss: 4.5376 - key_loss: 0.9417 - degree_1_loss: 0.2861 - degree_2_loss: 0.8939 - quality_loss: 0.7810 - inversion_loss: 0.6707 - root_loss: 0.9643 - key_accur 7/93 [=>............................] - ETA: 50s - loss: 4.5297 - key_loss: 0.9421 - degree_1_loss: 0.2905 - degree_2_loss: 0.8925 - quality_loss: 0.7766 - inversion_loss: 0.6745 - root_loss: 0.9535 - key_accur 8/93 [=>............................] - ETA: 49s - loss: 4.5284 - key_loss: 0.9418 - degree_1_loss: 0.2945 - degree_2_loss: 0.8922 - quality_loss: 0.7765 - inversion_loss: 0.6775 - root_loss: 0.9460 - key_accur 9/93 [=>............................] - ETA: 49s - loss: 4.5332 - key_loss: 0.9432 - degree_1_loss: 0.2975 - degree_2_loss: 0.8933 - quality_loss: 0.7768 - inversion_loss: 0.6800 - root_loss: 0.9424 - key_accur10/93 [==>...........................] - ETA: 48s - loss: 4.5299 - key_loss: 0.9423 - degree_1_loss: 0.3003 - degree_2_loss: 0.8932 - quality_loss: 0.7757 - inversion_loss: 0.6809 - root_loss: 0.9375 - key_accur11/93 [==>...........................] - ETA: 47s - loss: 4.5206 - key_loss: 0.9412 - degree_1_loss: 0.3017 - degree_2_loss: 0.8924 - quality_loss: 0.7737 - inversion_loss: 0.6802 - root_loss: 0.9314 - key_accur12/93 [==>...........................] - ETA: 47s - loss: 4.5208 - key_loss: 0.9432 - degree_1_loss: 0.3031 - degree_2_loss: 0.8917 - quality_loss: 0.7734 - inversion_loss: 0.6799 - root_loss: 0.9295 - key_accur13/93 [===>..........................] - ETA: 46s - loss: 4.5246 - key_loss: 0.9465 - degree_1_loss: 0.3046 - degree_2_loss: 0.8914 - quality_loss: 0.7740 - inversion_loss: 0.6795 - root_loss: 0.9286 - key_accur14/93 [===>..........................] - ETA: 45s - loss: 4.5262 - key_loss: 0.9481 - degree_1_loss: 0.3056 - degree_2_loss: 0.8909 - quality_loss: 0.7750 - inversion_loss: 0.6788 - root_loss: 0.9277 - key_accur15/93 [===>..........................] - ETA: 45s - loss: 4.5337 - key_loss: 0.9507 - degree_1_loss: 0.3071 - degree_2_loss: 0.8918 - quality_loss: 0.7767 - inversion_loss: 0.6787 - root_loss: 0.9287 - key_accur16/93 [====>.........................] - ETA: 44s - loss: 4.5439 - key_loss: 0.9540 - degree_1_loss: 0.3087 - degree_2_loss: 0.8936 - quality_loss: 0.7785 - inversion_loss: 0.6789 - root_loss: 0.9303 - key_accur17/93 [====>.........................] - ETA: 44s - loss: 4.5570 - key_loss: 0.9589 - degree_1_loss: 0.3100 - degree_2_loss: 0.8953 - quality_loss: 0.7805 - inversion_loss: 0.6796 - root_loss: 0.9328 - key_accur18/93 [====>.........................] - ETA: 43s - loss: 4.5701 - key_loss: 0.9641 - degree_1_loss: 0.3111 - degree_2_loss: 0.8968 - quality_loss: 0.7822 - inversion_loss: 0.6801 - root_loss: 0.9358 - key_accur19/93 [=====>........................] - ETA: 43s - loss: 4.5808 - key_loss: 0.9686 - degree_1_loss: 0.3122 - degree_2_loss: 0.8978 - quality_loss: 0.7836 - inversion_loss: 0.6806 - root_loss: 0.9380 - key_accur20/93 [=====>........................] - ETA: 42s - loss: 4.5909 - key_loss: 0.9732 - degree_1_loss: 0.3134 - degree_2_loss: 0.8991 - quality_loss: 0.7847 - inversion_loss: 0.6807 - root_loss: 0.9399 - key_accur21/93 [=====>........................] - ETA: 42s - loss: 4.5991 - key_loss: 0.9771 - degree_1_loss: 0.3144 - degree_2_loss: 0.8999 - quality_loss: 0.7855 - inversion_loss: 0.6805 - root_loss: 0.9416 - key_accur22/93 [======>.......................] - ETA: 41s - loss: 4.6049 - key_loss: 0.9799 - degree_1_loss: 0.3156 - degree_2_loss: 0.9006 - quality_loss: 0.7860 - inversion_loss: 0.6802 - root_loss: 0.9426 - key_accur23/93 [======>.......................] - ETA: 40s - loss: 4.6094 - key_loss: 0.9823 - degree_1_loss: 0.3166 - degree_2_loss: 0.9009 - quality_loss: 0.7863 - inversion_loss: 0.6800 - root_loss: 0.9433 - key_accur24/93 [======>.......................] - ETA: 40s - loss: 4.6150 - key_loss: 0.9851 - degree_1_loss: 0.3176 - degree_2_loss: 0.9014 - quality_loss: 0.7867 - inversion_loss: 0.6798 - root_loss: 0.9444 - key_accur25/93 [=======>......................] - ETA: 39s - loss: 4.6204 - key_loss: 0.9877 - degree_1_loss: 0.3185 - degree_2_loss: 0.9018 - quality_loss: 0.7873 - inversion_loss: 0.6796 - root_loss: 0.9455 - key_accur26/93 [=======>......................] - ETA: 39s - loss: 4.6237 - key_loss: 0.9897 - degree_1_loss: 0.3193 - degree_2_loss: 0.9018 - quality_loss: 0.7876 - inversion_loss: 0.6793 - root_loss: 0.9461 - key_accur27/93 [=======>......................] - ETA: 38s - loss: 4.6259 - key_loss: 0.9913 - degree_1_loss: 0.3199 - degree_2_loss: 0.9017 - quality_loss: 0.7878 - inversion_loss: 0.6789 - root_loss: 0.9464 - key_accur28/93 [========>.....................] - ETA: 37s - loss: 4.6305 - key_loss: 0.9935 - degree_1_loss: 0.3205 - degree_2_loss: 0.9019 - quality_loss: 0.7883 - inversion_loss: 0.6787 - root_loss: 0.9475 - key_accur29/93 [========>.....................] - ETA: 37s - loss: 4.6356 - key_loss: 0.9958 - degree_1_loss: 0.3209 - degree_2_loss: 0.9023 - quality_loss: 0.7889 - inversion_loss: 0.6788 - root_loss: 0.9487 - key_accur30/93 [========>.....................] - ETA: 36s - loss: 4.6401 - key_loss: 0.9978 - degree_1_loss: 0.3214 - degree_2_loss: 0.9026 - quality_loss: 0.7894 - inversion_loss: 0.6790 - root_loss: 0.9499 - key_accur31/93 [=========>....................] - ETA: 36s - loss: 4.6446 - key_loss: 0.9999 - degree_1_loss: 0.3219 - degree_2_loss: 0.9028 - quality_loss: 0.7899 - inversion_loss: 0.6791 - root_loss: 0.9510 - key_accur32/93 [=========>....................] - ETA: 35s - loss: 4.6509 - key_loss: 1.0026 - degree_1_loss: 0.3223 - degree_2_loss: 0.9034 - quality_loss: 0.7907 - inversion_loss: 0.6795 - root_loss: 0.9525 - key_accur33/93 [=========>....................] - ETA: 35s - loss: 4.6569 - key_loss: 1.0050 - degree_1_loss: 0.3228 - degree_2_loss: 0.9039 - quality_loss: 0.7915 - inversion_loss: 0.6799 - root_loss: 0.9538 - key_accur34/93 [=========>....................] - ETA: 34s - loss: 4.6627 - key_loss: 1.0074 - degree_1_loss: 0.3235 - degree_2_loss: 0.9044 - quality_loss: 0.7922 - inversion_loss: 0.6803 - root_loss: 0.9550 - key_accur35/93 [==========>...................] - ETA: 33s - loss: 4.6682 - key_loss: 1.0095 - degree_1_loss: 0.3242 - degree_2_loss: 0.9049 - quality_loss: 0.7928 - inversion_loss: 0.6806 - root_loss: 0.9561 - key_accur36/93 [==========>...................] - ETA: 33s - loss: 4.6738 - key_loss: 1.0117 - degree_1_loss: 0.3249 - degree_2_loss: 0.9054 - quality_loss: 0.7934 - inversion_loss: 0.6810 - root_loss: 0.9574 - key_accur37/93 [==========>...................] - ETA: 32s - loss: 4.6790 - key_loss: 1.0137 - degree_1_loss: 0.3256 - degree_2_loss: 0.9059 - quality_loss: 0.7939 - inversion_loss: 0.6814 - root_loss: 0.9586 - key_accur38/93 [===========>..................] - ETA: 32s - loss: 4.6843 - key_loss: 1.0156 - degree_1_loss: 0.3262 - degree_2_loss: 0.9064 - quality_loss: 0.7945 - inversion_loss: 0.6818 - root_loss: 0.9599 - key_accur39/93 [===========>..................] - ETA: 31s - loss: 4.6888 - key_loss: 1.0173 - degree_1_loss: 0.3268 - degree_2_loss: 0.9067 - quality_loss: 0.7950 - inversion_loss: 0.6821 - root_loss: 0.9609 - key_accur40/93 [===========>..................] - ETA: 31s - loss: 4.6942 - key_loss: 1.0193 - degree_1_loss: 0.3274 - degree_2_loss: 0.9072 - quality_loss: 0.7956 - inversion_loss: 0.6826 - root_loss: 0.9621 - key_accur41/93 [============>.................] - ETA: 30s - loss: 4.6985 - key_loss: 1.0210 - degree_1_loss: 0.3279 - degree_2_loss: 0.9076 - quality_loss: 0.7960 - inversion_loss: 0.6830 - root_loss: 0.9630 - key_accur42/93 [============>.................] - ETA: 29s - loss: 4.7029 - key_loss: 1.0225 - degree_1_loss: 0.3285 - degree_2_loss: 0.9081 - quality_loss: 0.7965 - inversion_loss: 0.6833 - root_loss: 0.9639 - key_accur43/93 [============>.................] - ETA: 29s - loss: 4.7078 - key_loss: 1.0241 - degree_1_loss: 0.3291 - degree_2_loss: 0.9086 - quality_loss: 0.7972 - inversion_loss: 0.6837 - root_loss: 0.9651 - key_accur44/93 [=============>................] - ETA: 28s - loss: 4.7124 - key_loss: 1.0255 - degree_1_loss: 0.3296 - degree_2_loss: 0.9092 - quality_loss: 0.7978 - inversion_loss: 0.6841 - root_loss: 0.9662 - key_accur45/93 [=============>................] - ETA: 28s - loss: 4.7169 - key_loss: 1.0270 - degree_1_loss: 0.3302 - degree_2_loss: 0.9097 - quality_loss: 0.7984 - inversion_loss: 0.6844 - root_loss: 0.9673 - key_accur46/93 [=============>................] - ETA: 27s - loss: 4.7212 - key_loss: 1.0285 - degree_1_loss: 0.3308 - degree_2_loss: 0.9101 - quality_loss: 0.7989 - inversion_loss: 0.6847 - root_loss: 0.9682 - key_accur47/93 [==============>...............] - ETA: 26s - loss: 4.7256 - key_loss: 1.0298 - degree_1_loss: 0.3313 - degree_2_loss: 0.9107 - quality_loss: 0.7995 - inversion_loss: 0.6850 - root_loss: 0.9692 - key_accur48/93 [==============>...............] - ETA: 26s - loss: 4.7300 - key_loss: 1.0313 - degree_1_loss: 0.3318 - degree_2_loss: 0.9112 - quality_loss: 0.8001 - inversion_loss: 0.6853 - root_loss: 0.9702 - key_accur49/93 [==============>...............] - ETA: 25s - loss: 4.7347 - key_loss: 1.0330 - degree_1_loss: 0.3323 - degree_2_loss: 0.9117 - quality_loss: 0.8008 - inversion_loss: 0.6856 - root_loss: 0.9713 - key_accur50/93 [===============>..............] - ETA: 25s - loss: 4.7390 - key_loss: 1.0345 - degree_1_loss: 0.3327 - degree_2_loss: 0.9122 - quality_loss: 0.8014 - inversion_loss: 0.6859 - root_loss: 0.9723 - key_accur51/93 [===============>..............] - ETA: 24s - loss: 4.7430 - key_loss: 1.0359 - degree_1_loss: 0.3331 - degree_2_loss: 0.9127 - quality_loss: 0.8020 - inversion_loss: 0.6861 - root_loss: 0.9732 - key_accur52/93 [===============>..............] - ETA: 24s - loss: 4.7471 - key_loss: 1.0373 - degree_1_loss: 0.3336 - degree_2_loss: 0.9132 - quality_loss: 0.8026 - inversion_loss: 0.6864 - root_loss: 0.9741 - key_accur53/93 [================>.............] - ETA: 23s - loss: 4.7507 - key_loss: 1.0385 - degree_1_loss: 0.3340 - degree_2_loss: 0.9136 - quality_loss: 0.8031 - inversion_loss: 0.6867 - root_loss: 0.9748 - key_accur54/93 [================>.............] - ETA: 22s - loss: 4.7543 - key_loss: 1.0397 - degree_1_loss: 0.3345 - degree_2_loss: 0.9140 - quality_loss: 0.8036 - inversion_loss: 0.6869 - root_loss: 0.9756 - key_accur55/93 [================>.............] - ETA: 22s - loss: 4.7578 - key_loss: 1.0410 - degree_1_loss: 0.3349 - degree_2_loss: 0.9143 - quality_loss: 0.8042 - inversion_loss: 0.6871 - root_loss: 0.9763 - key_accur56/93 [=================>............] - ETA: 21s - loss: 4.7611 - key_loss: 1.0421 - degree_1_loss: 0.3353 - degree_2_loss: 0.9146 - quality_loss: 0.8047 - inversion_loss: 0.6873 - root_loss: 0.9771 - key_accur57/93 [=================>............] - ETA: 21s - loss: 4.7641 - key_loss: 1.0432 - degree_1_loss: 0.3357 - degree_2_loss: 0.9150 - quality_loss: 0.8052 - inversion_loss: 0.6874 - root_loss: 0.9777 - key_accur58/93 [=================>............] - ETA: 20s - loss: 4.7673 - key_loss: 1.0444 - degree_1_loss: 0.3361 - degree_2_loss: 0.9153 - quality_loss: 0.8056 - inversion_loss: 0.6876 - root_loss: 0.9784 - key_accur59/93 [==================>...........] - ETA: 19s - loss: 4.7706 - key_loss: 1.0456 - degree_1_loss: 0.3364 - degree_2_loss: 0.9156 - quality_loss: 0.8061 - inversion_loss: 0.6877 - root_loss: 0.9792 - key_accur60/93 [==================>...........] - ETA: 19s - loss: 4.7744 - key_loss: 1.0469 - degree_1_loss: 0.3367 - degree_2_loss: 0.9160 - quality_loss: 0.8067 - inversion_loss: 0.6879 - root_loss: 0.9801 - key_accur61/93 [==================>...........] - ETA: 18s - loss: 4.7776 - key_loss: 1.0481 - degree_1_loss: 0.3371 - degree_2_loss: 0.9163 - quality_loss: 0.8071 - inversion_loss: 0.6881 - root_loss: 0.9809 - key_accur62/93 [===================>..........] - ETA: 18s - loss: 4.7806 - key_loss: 1.0492 - degree_1_loss: 0.3374 - degree_2_loss: 0.9166 - quality_loss: 0.8076 - inversion_loss: 0.6882 - root_loss: 0.9816 - key_accur63/93 [===================>..........] - ETA: 17s - loss: 4.7836 - key_loss: 1.0503 - degree_1_loss: 0.3377 - degree_2_loss: 0.9170 - quality_loss: 0.8080 - inversion_loss: 0.6883 - root_loss: 0.9824 - key_accur64/93 [===================>..........] - ETA: 17s - loss: 4.7865 - key_loss: 1.0513 - degree_1_loss: 0.3380 - degree_2_loss: 0.9172 - quality_loss: 0.8084 - inversion_loss: 0.6885 - root_loss: 0.9831 - key_accur65/93 [===================>..........] - ETA: 16s - loss: 4.7894 - key_loss: 1.0523 - degree_1_loss: 0.3383 - degree_2_loss: 0.9175 - quality_loss: 0.8088 - inversion_loss: 0.6886 - root_loss: 0.9839 - key_accur66/93 [====================>.........] - ETA: 15s - loss: 4.7921 - key_loss: 1.0533 - degree_1_loss: 0.3386 - degree_2_loss: 0.9178 - quality_loss: 0.8092 - inversion_loss: 0.6887 - root_loss: 0.9846 - key_accur67/93 [====================>.........] - ETA: 15s - loss: 4.7948 - key_loss: 1.0542 - degree_1_loss: 0.3388 - degree_2_loss: 0.9180 - quality_loss: 0.8096 - inversion_loss: 0.6888 - root_loss: 0.9852 - key_accur68/93 [====================>.........] - ETA: 14s - loss: 4.7972 - key_loss: 1.0551 - degree_1_loss: 0.3391 - degree_2_loss: 0.9183 - quality_loss: 0.8099 - inversion_loss: 0.6889 - root_loss: 0.9859 - key_accur69/93 [=====================>........] - ETA: 14s - loss: 4.7996 - key_loss: 1.0559 - degree_1_loss: 0.3393 - degree_2_loss: 0.9185 - quality_loss: 0.8103 - inversion_loss: 0.6890 - root_loss: 0.9865 - key_accur70/93 [=====================>........] - ETA: 13s - loss: 4.8018 - key_loss: 1.0568 - degree_1_loss: 0.3396 - degree_2_loss: 0.9187 - quality_loss: 0.8106 - inversion_loss: 0.6890 - root_loss: 0.9871 - key_accur71/93 [=====================>........] - ETA: 12s - loss: 4.8041 - key_loss: 1.0576 - degree_1_loss: 0.3398 - degree_2_loss: 0.9190 - quality_loss: 0.8109 - inversion_loss: 0.6890 - root_loss: 0.9878 - key_accur72/93 [======================>.......] - ETA: 12s - loss: 4.8063 - key_loss: 1.0584 - degree_1_loss: 0.3400 - degree_2_loss: 0.9191 - quality_loss: 0.8113 - inversion_loss: 0.6891 - root_loss: 0.9884 - key_accur73/93 [======================>.......] - ETA: 11s - loss: 4.8082 - key_loss: 1.0591 - degree_1_loss: 0.3402 - degree_2_loss: 0.9193 - quality_loss: 0.8116 - inversion_loss: 0.6891 - root_loss: 0.9889 - key_accur74/93 [======================>.......] - ETA: 11s - loss: 4.8102 - key_loss: 1.0599 - degree_1_loss: 0.3404 - degree_2_loss: 0.9195 - quality_loss: 0.8119 - inversion_loss: 0.6891 - root_loss: 0.9894 - key_accur75/93 [=======================>......] - ETA: 10s - loss: 4.8119 - key_loss: 1.0606 - degree_1_loss: 0.3406 - degree_2_loss: 0.9196 - quality_loss: 0.8122 - inversion_loss: 0.6891 - root_loss: 0.9899 - key_accur76/93 [=======================>......] - ETA: 9s - loss: 4.8137 - key_loss: 1.0613 - degree_1_loss: 0.3408 - degree_2_loss: 0.9198 - quality_loss: 0.8124 - inversion_loss: 0.6891 - root_loss: 0.9903 - key_accura77/93 [=======================>......] - ETA: 9s - loss: 4.8153 - key_loss: 1.0619 - degree_1_loss: 0.3410 - degree_2_loss: 0.9199 - quality_loss: 0.8127 - inversion_loss: 0.6891 - root_loss: 0.9908 - key_accura78/93 [========================>.....] - ETA: 8s - loss: 4.8168 - key_loss: 1.0626 - degree_1_loss: 0.3412 - degree_2_loss: 0.9200 - quality_loss: 0.8129 - inversion_loss: 0.6891 - root_loss: 0.9911 - key_accura79/93 [========================>.....] - ETA: 8s - loss: 4.8182 - key_loss: 1.0631 - degree_1_loss: 0.3414 - degree_2_loss: 0.9201 - quality_loss: 0.8131 - inversion_loss: 0.6891 - root_loss: 0.9914 - key_accura80/93 [========================>.....] - ETA: 7s - loss: 4.8199 - key_loss: 1.0638 - degree_1_loss: 0.3416 - degree_2_loss: 0.9202 - quality_loss: 0.8134 - inversion_loss: 0.6891 - root_loss: 0.9918 - key_accura81/93 [=========================>....] - ETA: 7s - loss: 4.8216 - key_loss: 1.0645 - degree_1_loss: 0.3418 - degree_2_loss: 0.9203 - quality_loss: 0.8136 - inversion_loss: 0.6891 - root_loss: 0.9922 - key_accura82/93 [=========================>....] - ETA: 6s - loss: 4.8236 - key_loss: 1.0653 - degree_1_loss: 0.3421 - degree_2_loss: 0.9204 - quality_loss: 0.8140 - inversion_loss: 0.6892 - root_loss: 0.9926 - key_accura83/93 [=========================>....] - ETA: 5s - loss: 4.8256 - key_loss: 1.0661 - degree_1_loss: 0.3423 - degree_2_loss: 0.9205 - quality_loss: 0.8143 - inversion_loss: 0.6893 - root_loss: 0.9931 - key_accura84/93 [==========================>...] - ETA: 5s - loss: 4.8275 - key_loss: 1.0669 - degree_1_loss: 0.3425 - degree_2_loss: 0.9207 - quality_loss: 0.8146 - inversion_loss: 0.6893 - root_loss: 0.9935 - key_accura85/93 [==========================>...] - ETA: 4s - loss: 4.8292 - key_loss: 1.0675 - degree_1_loss: 0.3427 - degree_2_loss: 0.9208 - quality_loss: 0.8149 - inversion_loss: 0.6893 - root_loss: 0.9940 - key_accura86/93 [==========================>...] - ETA: 4s - loss: 4.8309 - key_loss: 1.0682 - degree_1_loss: 0.3429 - degree_2_loss: 0.9209 - quality_loss: 0.8152 - inversion_loss: 0.6894 - root_loss: 0.9944 - key_accura87/93 [===========================>..] - ETA: 3s - loss: 4.8327 - key_loss: 1.0688 - degree_1_loss: 0.3431 - degree_2_loss: 0.9211 - quality_loss: 0.8155 - inversion_loss: 0.6894 - root_loss: 0.9948 - key_accura88/93 [===========================>..] - ETA: 2s - loss: 4.8345 - key_loss: 1.0694 - degree_1_loss: 0.3433 - degree_2_loss: 0.9212 - quality_loss: 0.8158 - inversion_loss: 0.6895 - root_loss: 0.9953 - key_accura89/93 [===========================>..] - ETA: 2s - loss: 4.8366 - key_loss: 1.0701 - degree_1_loss: 0.3435 - degree_2_loss: 0.9214 - quality_loss: 0.8161 - inversion_loss: 0.6896 - root_loss: 0.9959 - key_accura90/93 [============================>.] - ETA: 1s - loss: 4.8388 - key_loss: 1.0709 - degree_1_loss: 0.3437 - degree_2_loss: 0.9216 - quality_loss: 0.8165 - inversion_loss: 0.6896 - root_loss: 0.9965 - key_accura91/93 [============================>.] - ETA: 1s - loss: 4.8407 - key_loss: 1.0716 - degree_1_loss: 0.3438 - degree_2_loss: 0.9217 - quality_loss: 0.8169 - inversion_loss: 0.6897 - root_loss: 0.9970 - key_accura92/93 [============================>.] - ETA: 0s - loss: 4.8427 - key_loss: 1.0722 - degree_1_loss: 0.3440 - degree_2_loss: 0.9219 - quality_loss: 0.8172 - inversion_loss: 0.6898 - root_loss: 0.9976 - key_accura93/93 [==============================] - ETA: 0s - loss: 4.8446 - key_loss: 1.0729 - degree_1_loss: 0.3442 - degree_2_loss: 0.9221 - quality_loss: 0.8175 - inversion_loss: 0.6898 - root_loss: 0.9981 - key_accura93/93 [==============================] - 55s 590ms/step - loss: 4.8465 - key_loss: 1.0735 - degree_1_loss: 0.3443 - degree_2_loss: 0.9223 - quality_loss: 0.8179 - inversion_loss: 0.6899 - root_loss: 0.9987 - key_accuracy: 0.7140 - degree_1_accuracy: 0.9158 - degree_2_accuracy: 0.6946 - quality_accuracy: 0.7306 - inversion_accuracy: 0.7150 - root_accuracy: 0.7356 - val_loss: 6.9046 - val_key_loss: 1.7226 - val_degree_1_loss: 0.3614 - val_degree_2_loss: 1.1695 - val_quality_loss: 1.1990 - val_inversion_loss: 0.7560 - val_root_loss: 1.6960 - val_key_accuracy: 0.5433 - val_degree_1_accuracy: 0.9216 - val_degree_2_accuracy: 0.5844 - val_quality_accuracy: 0.6098 - val_inversion_accuracy: 0.7023 - val_root_accuracy: 0.5471 loss: 4.8465 key_loss: 1.0735 degree_1_loss: 0.3443 degree_2_loss: 0.9223 quality_loss: 0.8179 inversion_loss: 0.6899 root_loss: 0.9987 key_accuracy: 0.7140 degree_1_accuracy: 0.9158 degree_2_accuracy: 0.6946 quality_accuracy: 0.7306 inversion_accuracy: 0.7150 root_accuracy: 0.7356 val_loss: 6.9046 val_key_loss: 1.7226 val_degree_1_loss: 0.3614 val_degree_2_loss: 1.1695 val_quality_loss: 1.1990 val_inversion_loss: 0.7560 val_root_loss: 1.6960 val_key_accuracy: 0.5433 val_degree_1_accuracy: 0.9216 val_degree_2_accuracy: 0.5844 val_quality_accuracy: 0.6098 val_inversion_accuracy: 0.7023 val_root_accuracy: 0.5471 ``` The output of `analyse_results`: ```python {'d7 no inv': 6.7765567765567765, 'degree': 54.73046903494055, 'degree 1': 92.15843606389963, 'degree 2': 58.43606389962798, 'derived root': 46.130279378510465, 'inversion': 70.23123495513896, 'key': 54.329272740535416, 'quality': 60.981836749580566, 'roman': 35.16667882413013, 'roman + inv': 27.222992194908453, 'root': 54.70858560070027, 'root coherence': 72.7186519804508, 'secondary': 0.0} ```

The output on BPSSynth seems off. It may be off or it may be a misalignment problem produced at preprocessing.py. Either way, I will assume for now that the results are correct.

UPDATE: It was off. I set the code to generate BPSSynth to explicitly .expandRepeats() before generating score and csv, and that seems to have corrected the issue.

Buggy version of experiment "BPSSynth on BPS-val", before fixing issue with `expandRepeats()`

Here is `BPSSynth` on the training and real `BPS` on validation `spelling_bass_cut` and `conv_gru` as with the previous experiments. ```python Epoch 18/100 1/93 [..............................] - ETA: 1:03 - loss: 4.4423 - key_loss: 0.7465 - degree_1_loss: 0.4513 - degree_2_loss: 0.8506 - quality_loss: 0.7579 - inversion_loss: 0.6197 - root_loss: 1.0164 - key_accu 2/93 [..............................] - ETA: 53s - loss: 4.3277 - key_loss: 0.7724 - degree_1_loss: 0.4217 - degree_2_loss: 0.8533 - quality_loss: 0.7490 - inversion_loss: 0.6007 - root_loss: 0.9306 - key_accur 3/93 [..............................] - ETA: 52s - loss: 4.2079 - key_loss: 0.7691 - degree_1_loss: 0.4054 - degree_2_loss: 0.8423 - quality_loss: 0.7338 - inversion_loss: 0.5828 - root_loss: 0.8746 - key_accur 4/93 [>.............................] - ETA: 51s - loss: 4.1653 - key_loss: 0.7703 - degree_1_loss: 0.3923 - degree_2_loss: 0.8406 - quality_loss: 0.7250 - inversion_loss: 0.5785 - root_loss: 0.8586 - key_accur 5/93 [>.............................] - ETA: 51s - loss: 4.1268 - key_loss: 0.7780 - degree_1_loss: 0.3887 - degree_2_loss: 0.8350 - quality_loss: 0.7153 - inversion_loss: 0.5713 - root_loss: 0.8386 - key_accur 6/93 [>.............................] - ETA: 50s - loss: 4.1611 - key_loss: 0.8045 - degree_1_loss: 0.3861 - degree_2_loss: 0.8370 - quality_loss: 0.7184 - inversion_loss: 0.5722 - root_loss: 0.8429 - key_accur 7/93 [=>............................] - ETA: 50s - loss: 4.1784 - key_loss: 0.8176 - degree_1_loss: 0.3842 - degree_2_loss: 0.8383 - quality_loss: 0.7203 - inversion_loss: 0.5723 - root_loss: 0.8457 - key_accur 8/93 [=>............................] - ETA: 49s - loss: 4.2176 - key_loss: 0.8351 - degree_1_loss: 0.3830 - degree_2_loss: 0.8432 - quality_loss: 0.7261 - inversion_loss: 0.5744 - root_loss: 0.8558 - key_accur 9/93 [=>............................] - ETA: 49s - loss: 4.2472 - key_loss: 0.8465 - degree_1_loss: 0.3812 - degree_2_loss: 0.8473 - quality_loss: 0.7312 - inversion_loss: 0.5767 - root_loss: 0.8643 - key_accur10/93 [==>...........................] - ETA: 48s - loss: 4.2813 - key_loss: 0.8577 - degree_1_loss: 0.3800 - degree_2_loss: 0.8522 - quality_loss: 0.7375 - inversion_loss: 0.5798 - root_loss: 0.8741 - key_accur11/93 [==>...........................] - ETA: 48s - loss: 4.3143 - key_loss: 0.8708 - degree_1_loss: 0.3792 - degree_2_loss: 0.8559 - quality_loss: 0.7426 - inversion_loss: 0.5822 - root_loss: 0.8836 - key_accur12/93 [==>...........................] - ETA: 47s - loss: 4.3486 - key_loss: 0.8853 - degree_1_loss: 0.3781 - degree_2_loss: 0.8590 - quality_loss: 0.7475 - inversion_loss: 0.5852 - root_loss: 0.8936 - key_accur13/93 [===>..........................] - ETA: 46s - loss: 4.3749 - key_loss: 0.8966 - degree_1_loss: 0.3770 - degree_2_loss: 0.8610 - quality_loss: 0.7515 - inversion_loss: 0.5877 - root_loss: 0.9012 - key_accur14/93 [===>..........................] - ETA: 46s - loss: 4.3965 - key_loss: 0.9060 - degree_1_loss: 0.3759 - degree_2_loss: 0.8626 - quality_loss: 0.7547 - inversion_loss: 0.5897 - root_loss: 0.9075 - key_accur15/93 [===>..........................] - ETA: 45s - loss: 4.4137 - key_loss: 0.9138 - degree_1_loss: 0.3748 - degree_2_loss: 0.8643 - quality_loss: 0.7570 - inversion_loss: 0.5913 - root_loss: 0.9125 - key_accur16/93 [====>.........................] - ETA: 45s - loss: 4.4305 - key_loss: 0.9213 - degree_1_loss: 0.3742 - degree_2_loss: 0.8655 - quality_loss: 0.7590 - inversion_loss: 0.5928 - root_loss: 0.9178 - key_accur17/93 [====>.........................] - ETA: 44s - loss: 4.4450 - key_loss: 0.9277 - degree_1_loss: 0.3736 - degree_2_loss: 0.8666 - quality_loss: 0.7606 - inversion_loss: 0.5940 - root_loss: 0.9224 - key_accur18/93 [====>.........................] - ETA: 44s - loss: 4.4543 - key_loss: 0.9330 - degree_1_loss: 0.3729 - degree_2_loss: 0.8668 - quality_loss: 0.7613 - inversion_loss: 0.5946 - root_loss: 0.9257 - key_accur19/93 [=====>........................] - ETA: 43s - loss: 4.4624 - key_loss: 0.9374 - degree_1_loss: 0.3725 - degree_2_loss: 0.8669 - quality_loss: 0.7620 - inversion_loss: 0.5952 - root_loss: 0.9283 - key_accur20/93 [=====>........................] - ETA: 42s - loss: 4.4695 - key_loss: 0.9416 - degree_1_loss: 0.3722 - degree_2_loss: 0.8671 - quality_loss: 0.7624 - inversion_loss: 0.5955 - root_loss: 0.9307 - key_accur21/93 [=====>........................] - ETA: 42s - loss: 4.4746 - key_loss: 0.9453 - degree_1_loss: 0.3718 - degree_2_loss: 0.8672 - quality_loss: 0.7626 - inversion_loss: 0.5955 - root_loss: 0.9322 - key_accur22/93 [======>.......................] - ETA: 41s - loss: 4.4812 - key_loss: 0.9486 - degree_1_loss: 0.3714 - degree_2_loss: 0.8679 - quality_loss: 0.7634 - inversion_loss: 0.5958 - root_loss: 0.9341 - key_accur23/93 [======>.......................] - ETA: 41s - loss: 4.4880 - key_loss: 0.9513 - degree_1_loss: 0.3710 - degree_2_loss: 0.8690 - quality_loss: 0.7644 - inversion_loss: 0.5962 - root_loss: 0.9362 - key_accur24/93 [======>.......................] - ETA: 40s - loss: 4.4967 - key_loss: 0.9544 - degree_1_loss: 0.3707 - degree_2_loss: 0.8703 - quality_loss: 0.7658 - inversion_loss: 0.5966 - root_loss: 0.9389 - key_accur25/93 [=======>......................] - ETA: 40s - loss: 4.5027 - key_loss: 0.9566 - degree_1_loss: 0.3701 - degree_2_loss: 0.8713 - quality_loss: 0.7669 - inversion_loss: 0.5969 - root_loss: 0.9409 - key_accur26/93 [=======>......................] - ETA: 39s - loss: 4.5088 - key_loss: 0.9591 - degree_1_loss: 0.3697 - degree_2_loss: 0.8722 - quality_loss: 0.7678 - inversion_loss: 0.5971 - root_loss: 0.9428 - key_accur27/93 [=======>......................] - ETA: 38s - loss: 4.5138 - key_loss: 0.9613 - degree_1_loss: 0.3692 - degree_2_loss: 0.8729 - quality_loss: 0.7686 - inversion_loss: 0.5973 - root_loss: 0.9444 - key_accur28/93 [========>.....................] - ETA: 38s - loss: 4.5182 - key_loss: 0.9632 - degree_1_loss: 0.3689 - degree_2_loss: 0.8735 - quality_loss: 0.7693 - inversion_loss: 0.5975 - root_loss: 0.9457 - key_accur29/93 [========>.....................] - ETA: 37s - loss: 4.5213 - key_loss: 0.9647 - degree_1_loss: 0.3685 - degree_2_loss: 0.8739 - quality_loss: 0.7698 - inversion_loss: 0.5975 - root_loss: 0.9468 - key_accur30/93 [========>.....................] - ETA: 37s - loss: 4.5246 - key_loss: 0.9662 - degree_1_loss: 0.3682 - degree_2_loss: 0.8745 - quality_loss: 0.7704 - inversion_loss: 0.5975 - root_loss: 0.9478 - key_accur31/93 [=========>....................] - ETA: 36s - loss: 4.5282 - key_loss: 0.9678 - degree_1_loss: 0.3680 - degree_2_loss: 0.8750 - quality_loss: 0.7710 - inversion_loss: 0.5976 - root_loss: 0.9487 - key_accur32/93 [=========>....................] - ETA: 36s - loss: 4.5310 - key_loss: 0.9692 - degree_1_loss: 0.3678 - degree_2_loss: 0.8755 - quality_loss: 0.7714 - inversion_loss: 0.5977 - root_loss: 0.9494 - key_accur33/93 [=========>....................] - ETA: 35s - loss: 4.5332 - key_loss: 0.9706 - degree_1_loss: 0.3675 - degree_2_loss: 0.8759 - quality_loss: 0.7717 - inversion_loss: 0.5977 - root_loss: 0.9499 - key_accur34/93 [=========>....................] - ETA: 34s - loss: 4.5344 - key_loss: 0.9716 - degree_1_loss: 0.3672 - degree_2_loss: 0.8760 - quality_loss: 0.7718 - inversion_loss: 0.5976 - root_loss: 0.9501 - key_accur35/93 [==========>...................] - ETA: 34s - loss: 4.5358 - key_loss: 0.9726 - degree_1_loss: 0.3669 - degree_2_loss: 0.8762 - quality_loss: 0.7719 - inversion_loss: 0.5976 - root_loss: 0.9505 - key_accur36/93 [==========>...................] - ETA: 33s - loss: 4.5367 - key_loss: 0.9735 - degree_1_loss: 0.3665 - degree_2_loss: 0.8763 - quality_loss: 0.7720 - inversion_loss: 0.5976 - root_loss: 0.9508 - key_accur37/93 [==========>...................] - ETA: 33s - loss: 4.5365 - key_loss: 0.9741 - degree_1_loss: 0.3662 - degree_2_loss: 0.8762 - quality_loss: 0.7719 - inversion_loss: 0.5974 - root_loss: 0.9508 - key_accur38/93 [===========>..................] - ETA: 32s - loss: 4.5366 - key_loss: 0.9746 - degree_1_loss: 0.3658 - degree_2_loss: 0.8761 - quality_loss: 0.7718 - inversion_loss: 0.5974 - root_loss: 0.9508 - key_accur39/93 [===========>..................] - ETA: 31s - loss: 4.5374 - key_loss: 0.9752 - degree_1_loss: 0.3655 - degree_2_loss: 0.8762 - quality_loss: 0.7720 - inversion_loss: 0.5974 - root_loss: 0.9511 - key_accur40/93 [===========>..................] - ETA: 31s - loss: 4.5383 - key_loss: 0.9758 - degree_1_loss: 0.3653 - degree_2_loss: 0.8763 - quality_loss: 0.7721 - inversion_loss: 0.5975 - root_loss: 0.9514 - key_accur41/93 [============>.................] - ETA: 30s - loss: 4.5402 - key_loss: 0.9763 - degree_1_loss: 0.3652 - degree_2_loss: 0.8766 - quality_loss: 0.7724 - inversion_loss: 0.5977 - root_loss: 0.9521 - key_accur42/93 [============>.................] - ETA: 30s - loss: 4.5420 - key_loss: 0.9767 - degree_1_loss: 0.3650 - degree_2_loss: 0.8769 - quality_loss: 0.7727 - inversion_loss: 0.5979 - root_loss: 0.9527 - key_accur43/93 [============>.................] - ETA: 29s - loss: 4.5432 - key_loss: 0.9770 - degree_1_loss: 0.3649 - degree_2_loss: 0.8772 - quality_loss: 0.7729 - inversion_loss: 0.5981 - root_loss: 0.9531 - key_accur44/93 [=============>................] - ETA: 28s - loss: 4.5442 - key_loss: 0.9772 - degree_1_loss: 0.3647 - degree_2_loss: 0.8774 - quality_loss: 0.7731 - inversion_loss: 0.5983 - root_loss: 0.9535 - key_accur45/93 [=============>................] - ETA: 28s - loss: 4.5447 - key_loss: 0.9773 - degree_1_loss: 0.3645 - degree_2_loss: 0.8776 - quality_loss: 0.7731 - inversion_loss: 0.5984 - root_loss: 0.9538 - key_accur46/93 [=============>................] - ETA: 27s - loss: 4.5447 - key_loss: 0.9773 - degree_1_loss: 0.3643 - degree_2_loss: 0.8777 - quality_loss: 0.7731 - inversion_loss: 0.5985 - root_loss: 0.9539 - key_accur47/93 [==============>...............] - ETA: 27s - loss: 4.5450 - key_loss: 0.9773 - degree_1_loss: 0.3641 - degree_2_loss: 0.8777 - quality_loss: 0.7732 - inversion_loss: 0.5986 - root_loss: 0.9541 - key_accur48/93 [==============>...............] - ETA: 26s - loss: 4.5445 - key_loss: 0.9772 - degree_1_loss: 0.3639 - degree_2_loss: 0.8777 - quality_loss: 0.7731 - inversion_loss: 0.5986 - root_loss: 0.9541 - key_accur49/93 [==============>...............] - ETA: 26s - loss: 4.5443 - key_loss: 0.9770 - degree_1_loss: 0.3637 - degree_2_loss: 0.8777 - quality_loss: 0.7732 - inversion_loss: 0.5986 - root_loss: 0.9541 - key_accur50/93 [===============>..............] - ETA: 25s - loss: 4.5442 - key_loss: 0.9769 - degree_1_loss: 0.3636 - degree_2_loss: 0.8777 - quality_loss: 0.7732 - inversion_loss: 0.5986 - root_loss: 0.9542 - key_accur51/93 [===============>..............] - ETA: 24s - loss: 4.5439 - key_loss: 0.9768 - degree_1_loss: 0.3634 - degree_2_loss: 0.8776 - quality_loss: 0.7732 - inversion_loss: 0.5987 - root_loss: 0.9542 - key_accur52/93 [===============>..............] - ETA: 24s - loss: 4.5437 - key_loss: 0.9767 - degree_1_loss: 0.3633 - degree_2_loss: 0.8776 - quality_loss: 0.7732 - inversion_loss: 0.5987 - root_loss: 0.9543 - key_accur53/93 [================>.............] - ETA: 23s - loss: 4.5443 - key_loss: 0.9769 - degree_1_loss: 0.3632 - degree_2_loss: 0.8777 - quality_loss: 0.7733 - inversion_loss: 0.5987 - root_loss: 0.9546 - key_accur54/93 [================>.............] - ETA: 23s - loss: 4.5446 - key_loss: 0.9770 - degree_1_loss: 0.3630 - degree_2_loss: 0.8776 - quality_loss: 0.7733 - inversion_loss: 0.5987 - root_loss: 0.9549 - key_accur55/93 [================>.............] - ETA: 22s - loss: 4.5450 - key_loss: 0.9770 - degree_1_loss: 0.3629 - degree_2_loss: 0.8776 - quality_loss: 0.7734 - inversion_loss: 0.5987 - root_loss: 0.9552 - key_accur56/93 [=================>............] - ETA: 21s - loss: 4.5448 - key_loss: 0.9770 - degree_1_loss: 0.3628 - degree_2_loss: 0.8776 - quality_loss: 0.7734 - inversion_loss: 0.5987 - root_loss: 0.9554 - key_accur57/93 [=================>............] - ETA: 21s - loss: 4.5445 - key_loss: 0.9770 - degree_1_loss: 0.3626 - degree_2_loss: 0.8774 - quality_loss: 0.7734 - inversion_loss: 0.5986 - root_loss: 0.9555 - key_accur58/93 [=================>............] - ETA: 20s - loss: 4.5446 - key_loss: 0.9770 - degree_1_loss: 0.3624 - degree_2_loss: 0.8774 - quality_loss: 0.7735 - inversion_loss: 0.5985 - root_loss: 0.9557 - key_accur59/93 [==================>...........] - ETA: 20s - loss: 4.5446 - key_loss: 0.9772 - degree_1_loss: 0.3622 - degree_2_loss: 0.8773 - quality_loss: 0.7735 - inversion_loss: 0.5985 - root_loss: 0.9559 - key_accur60/93 [==================>...........] - ETA: 19s - loss: 4.5446 - key_loss: 0.9773 - degree_1_loss: 0.3620 - degree_2_loss: 0.8771 - quality_loss: 0.7736 - inversion_loss: 0.5984 - root_loss: 0.9561 - key_accur61/93 [==================>...........] - ETA: 18s - loss: 4.5444 - key_loss: 0.9773 - degree_1_loss: 0.3618 - degree_2_loss: 0.8770 - quality_loss: 0.7736 - inversion_loss: 0.5983 - root_loss: 0.9563 - key_accur62/93 [===================>..........] - ETA: 18s - loss: 4.5443 - key_loss: 0.9774 - degree_1_loss: 0.3616 - degree_2_loss: 0.8769 - quality_loss: 0.7736 - inversion_loss: 0.5982 - root_loss: 0.9565 - key_accur63/93 [===================>..........] - ETA: 17s - loss: 4.5439 - key_loss: 0.9774 - degree_1_loss: 0.3615 - degree_2_loss: 0.8768 - quality_loss: 0.7736 - inversion_loss: 0.5981 - root_loss: 0.9566 - key_accur64/93 [===================>..........] - ETA: 17s - loss: 4.5437 - key_loss: 0.9774 - degree_1_loss: 0.3613 - degree_2_loss: 0.8767 - quality_loss: 0.7736 - inversion_loss: 0.5979 - root_loss: 0.9567 - key_accur65/93 [===================>..........] - ETA: 16s - loss: 4.5437 - key_loss: 0.9776 - degree_1_loss: 0.3611 - degree_2_loss: 0.8766 - quality_loss: 0.7737 - inversion_loss: 0.5978 - root_loss: 0.9569 - key_accur66/93 [====================>.........] - ETA: 16s - loss: 4.5439 - key_loss: 0.9777 - degree_1_loss: 0.3609 - degree_2_loss: 0.8766 - quality_loss: 0.7737 - inversion_loss: 0.5978 - root_loss: 0.9572 - key_accur67/93 [====================>.........] - ETA: 15s - loss: 4.5441 - key_loss: 0.9779 - degree_1_loss: 0.3607 - degree_2_loss: 0.8765 - quality_loss: 0.7738 - inversion_loss: 0.5977 - root_loss: 0.9574 - key_accur68/93 [====================>.........] - ETA: 14s - loss: 4.5443 - key_loss: 0.9780 - degree_1_loss: 0.3606 - degree_2_loss: 0.8765 - quality_loss: 0.7739 - inversion_loss: 0.5977 - root_loss: 0.9577 - key_accur69/93 [=====================>........] - ETA: 14s - loss: 4.5444 - key_loss: 0.9781 - degree_1_loss: 0.3604 - degree_2_loss: 0.8764 - quality_loss: 0.7740 - inversion_loss: 0.5976 - root_loss: 0.9579 - key_accur70/93 [=====================>........] - ETA: 13s - loss: 4.5447 - key_loss: 0.9783 - degree_1_loss: 0.3602 - degree_2_loss: 0.8764 - quality_loss: 0.7741 - inversion_loss: 0.5975 - root_loss: 0.9582 - key_accur71/93 [=====================>........] - ETA: 13s - loss: 4.5452 - key_loss: 0.9785 - degree_1_loss: 0.3601 - degree_2_loss: 0.8764 - quality_loss: 0.7742 - inversion_loss: 0.5975 - root_loss: 0.9584 - key_accur72/93 [======================>.......] - ETA: 12s - loss: 4.5454 - key_loss: 0.9786 - degree_1_loss: 0.3599 - degree_2_loss: 0.8764 - quality_loss: 0.7744 - inversion_loss: 0.5975 - root_loss: 0.9587 - key_accur73/93 [======================>.......] - ETA: 11s - loss: 4.5455 - key_loss: 0.9787 - degree_1_loss: 0.3597 - degree_2_loss: 0.8763 - quality_loss: 0.7745 - inversion_loss: 0.5974 - root_loss: 0.9588 - key_accur74/93 [======================>.......] - ETA: 11s - loss: 4.5455 - key_loss: 0.9787 - degree_1_loss: 0.3596 - degree_2_loss: 0.8763 - quality_loss: 0.7745 - inversion_loss: 0.5973 - root_loss: 0.9590 - key_accur75/93 [=======================>......] - ETA: 10s - loss: 4.5456 - key_loss: 0.9788 - degree_1_loss: 0.3595 - degree_2_loss: 0.8762 - quality_loss: 0.7746 - inversion_loss: 0.5973 - root_loss: 0.9592 - key_accur76/93 [=======================>......] - ETA: 10s - loss: 4.5459 - key_loss: 0.9789 - degree_1_loss: 0.3593 - degree_2_loss: 0.8763 - quality_loss: 0.7747 - inversion_loss: 0.5972 - root_loss: 0.9595 - key_accur77/93 [=======================>......] - ETA: 9s - loss: 4.5462 - key_loss: 0.9790 - degree_1_loss: 0.3592 - degree_2_loss: 0.8763 - quality_loss: 0.7749 - inversion_loss: 0.5972 - root_loss: 0.9597 - key_accura78/93 [========================>.....] - ETA: 8s - loss: 4.5464 - key_loss: 0.9790 - degree_1_loss: 0.3591 - degree_2_loss: 0.8763 - quality_loss: 0.7750 - inversion_loss: 0.5972 - root_loss: 0.9599 - key_accura79/93 [========================>.....] - ETA: 8s - loss: 4.5466 - key_loss: 0.9791 - degree_1_loss: 0.3590 - degree_2_loss: 0.8762 - quality_loss: 0.7751 - inversion_loss: 0.5971 - root_loss: 0.9602 - key_accura80/93 [========================>.....] - ETA: 7s - loss: 4.5467 - key_loss: 0.9791 - degree_1_loss: 0.3589 - degree_2_loss: 0.8762 - quality_loss: 0.7751 - inversion_loss: 0.5970 - root_loss: 0.9604 - key_accura81/93 [=========================>....] - ETA: 7s - loss: 4.5468 - key_loss: 0.9792 - degree_1_loss: 0.3587 - degree_2_loss: 0.8762 - quality_loss: 0.7752 - inversion_loss: 0.5970 - root_loss: 0.9605 - key_accura82/93 [=========================>....] - ETA: 6s - loss: 4.5468 - key_loss: 0.9792 - degree_1_loss: 0.3586 - degree_2_loss: 0.8761 - quality_loss: 0.7752 - inversion_loss: 0.5969 - root_loss: 0.9607 - key_accura83/93 [=========================>....] - ETA: 5s - loss: 4.5469 - key_loss: 0.9793 - degree_1_loss: 0.3585 - degree_2_loss: 0.8761 - quality_loss: 0.7753 - inversion_loss: 0.5969 - root_loss: 0.9608 - key_accura84/93 [==========================>...] - ETA: 5s - loss: 4.5470 - key_loss: 0.9794 - degree_1_loss: 0.3584 - degree_2_loss: 0.8760 - quality_loss: 0.7754 - inversion_loss: 0.5968 - root_loss: 0.9610 - key_accura85/93 [==========================>...] - ETA: 4s - loss: 4.5469 - key_loss: 0.9794 - degree_1_loss: 0.3583 - degree_2_loss: 0.8760 - quality_loss: 0.7754 - inversion_loss: 0.5967 - root_loss: 0.9611 - key_accura86/93 [==========================>...] - ETA: 4s - loss: 4.5469 - key_loss: 0.9794 - degree_1_loss: 0.3582 - degree_2_loss: 0.8759 - quality_loss: 0.7755 - inversion_loss: 0.5967 - root_loss: 0.9613 - key_accura87/93 [===========================>..] - ETA: 3s - loss: 4.5469 - key_loss: 0.9794 - degree_1_loss: 0.3580 - degree_2_loss: 0.8759 - quality_loss: 0.7755 - inversion_loss: 0.5966 - root_loss: 0.9614 - key_accura88/93 [===========================>..] - ETA: 2s - loss: 4.5470 - key_loss: 0.9795 - degree_1_loss: 0.3579 - degree_2_loss: 0.8759 - quality_loss: 0.7756 - inversion_loss: 0.5966 - root_loss: 0.9616 - key_accura89/93 [===========================>..] - ETA: 2s - loss: 4.5472 - key_loss: 0.9797 - degree_1_loss: 0.3577 - degree_2_loss: 0.8758 - quality_loss: 0.7756 - inversion_loss: 0.5966 - root_loss: 0.9618 - key_accura90/93 [============================>.] - ETA: 1s - loss: 4.5472 - key_loss: 0.9798 - degree_1_loss: 0.3576 - degree_2_loss: 0.8758 - quality_loss: 0.7757 - inversion_loss: 0.5965 - root_loss: 0.9619 - key_accura91/93 [============================>.] - ETA: 1s - loss: 4.5472 - key_loss: 0.9798 - degree_1_loss: 0.3574 - degree_2_loss: 0.8757 - quality_loss: 0.7757 - inversion_loss: 0.5965 - root_loss: 0.9621 - key_accura92/93 [============================>.] - ETA: 0s - loss: 4.5473 - key_loss: 0.9799 - degree_1_loss: 0.3573 - degree_2_loss: 0.8757 - quality_loss: 0.7757 - inversion_loss: 0.5965 - root_loss: 0.9622 - key_accura93/93 [==============================] - ETA: 0s - loss: 4.5474 - key_loss: 0.9800 - degree_1_loss: 0.3571 - degree_2_loss: 0.8756 - quality_loss: 0.7758 - inversion_loss: 0.5965 - root_loss: 0.9624 - key_accura93/93 [==============================] - 56s 599ms/step - loss: 4.5475 - key_loss: 0.9801 - degree_1_loss: 0.3570 - degree_2_loss: 0.8756 - quality_loss: 0.7758 - inversion_loss: 0.5964 - root_loss: 0.9626 - key_accuracy: 0.7370 - degree_1_accuracy: 0.9108 - degree_2_accuracy: 0.7078 - quality_accuracy: 0.7509 - inversion_accuracy: 0.7774 - root_accuracy: 0.7446 - val_loss: 6.2563 - val_key_loss: 1.2258 - val_degree_1_loss: 0.4205 - val_degree_2_loss: 1.2745 - val_quality_loss: 1.2359 - val_inversion_loss: 0.9754 - val_root_loss: 1.1243 - val_key_accuracy: 0.6256 - val_degree_1_accuracy: 0.9080 - val_degree_2_accuracy: 0.5341 - val_quality_accuracy: 0.5986 - val_inversion_accuracy: 0.6141 - val_root_accuracy: 0.6498 loss: 4.5475 key_loss: 0.9801 degree_1_loss: 0.3570 degree_2_loss: 0.8756 quality_loss: 0.7758 inversion_loss: 0.5964 root_loss: 0.9626 key_accuracy: 0.7370 degree_1_accuracy: 0.9108 degree_2_accuracy: 0.7078 quality_accuracy: 0.7509 inversion_accuracy: 0.7774 root_accuracy: 0.7446 val_loss: 6.2563 val_key_loss: 1.2258 val_degree_1_loss: 0.4205 val_degree_2_loss: 1.2745 val_quality_loss: 1.2359 val_inversion_loss: 0.9754 val_root_loss: 1.1243 val_key_accuracy: 0.6256 val_degree_1_accuracy: 0.9080 val_degree_2_accuracy: 0.5341 val_quality_accuracy: 0.5986 val_inversion_accuracy: 0.6141 val_root_accuracy: 0.6498 ``` The output of `analyse_results`: ```python {'key': 62.55755139096038, 'degree 1': 90.79826211010959, 'degree 2': 53.40769081123144, 'quality': 59.85993126256404, 'inversion': 61.4097659036379, 'root': 64.98281564100901, 'degree': 49.41313792879839, 'secondary': 0.0, 'derived root': 46.22916801763829, 'roman': 29.680306076129952, 'roman + inv': 18.889825562544583, 'root coherence': 53.77083198236171, 'd7 no inv': 3.524804177545692} ```

Buggy version of experiment "BPSSynth+BPS on BPS-val", before fixing issue with `expandRepeats()`

Here is `BPS + BPSSynth` on training and `BPS` on validation. `spelling_bass_cut` and `conv_gru` as usual: ```python Epoch 19/100 1/197 [..............................] - ETA: 2:26 - loss: 3.8030 - key_loss: 0.5788 - degree_1_loss: 0.2730 - degree_2_loss: 0.7603 - quality_loss: 0.7172 - inversion_loss: 0.6279 - root_loss: 0.8458 - key_ac 2/197 [..............................] - ETA: 1:49 - loss: 4.1334 - key_loss: 0.6812 - degree_1_loss: 0.2992 - degree_2_loss: 0.8020 - quality_loss: 0.7715 - inversion_loss: 0.6702 - root_loss: 0.9092 - key_ac 3/197 [..............................] - ETA: 1:49 - loss: 4.4856 - key_loss: 0.7981 - degree_1_loss: 0.3137 - degree_2_loss: 0.8545 - quality_loss: 0.8263 - inversion_loss: 0.7157 - root_loss: 0.9773 - key_ac 4/197 [..............................] - ETA: 1:49 - loss: 4.6141 - key_loss: 0.8389 - degree_1_loss: 0.3218 - degree_2_loss: 0.8687 - quality_loss: 0.8449 - inversion_loss: 0.7440 - root_loss: 0.9957 - key_ac 5/197 [..............................] - ETA: 1:49 - loss: 4.6079 - key_loss: 0.8428 - degree_1_loss: 0.3239 - degree_2_loss: 0.8637 - quality_loss: 0.8421 - inversion_loss: 0.7469 - root_loss: 0.9885 - key_ac 6/197 [..............................] - ETA: 1:49 - loss: 4.5748 - key_loss: 0.8355 - degree_1_loss: 0.3250 - degree_2_loss: 0.8554 - quality_loss: 0.8369 - inversion_loss: 0.7459 - root_loss: 0.9761 - key_ac 7/197 [>.............................] - ETA: 1:49 - loss: 4.5504 - key_loss: 0.8282 - degree_1_loss: 0.3249 - degree_2_loss: 0.8494 - quality_loss: 0.8349 - inversion_loss: 0.7442 - root_loss: 0.9689 - key_ac 8/197 [>.............................] - ETA: 1:48 - loss: 4.5239 - key_loss: 0.8173 - degree_1_loss: 0.3259 - degree_2_loss: 0.8452 - quality_loss: 0.8316 - inversion_loss: 0.7425 - root_loss: 0.9614 - key_ac 9/197 [>.............................] - ETA: 1:49 - loss: 4.5122 - key_loss: 0.8099 - degree_1_loss: 0.3263 - degree_2_loss: 0.8436 - quality_loss: 0.8312 - inversion_loss: 0.7419 - root_loss: 0.9593 - key_ac 10/197 [>.............................] - ETA: 1:48 - loss: 4.4924 - key_loss: 0.8026 - degree_1_loss: 0.3259 - degree_2_loss: 0.8402 - quality_loss: 0.8292 - inversion_loss: 0.7397 - root_loss: 0.9549 - key_ac 11/197 [>.............................] - ETA: 1:47 - loss: 4.4709 - key_loss: 0.7954 - degree_1_loss: 0.3261 - degree_2_loss: 0.8377 - quality_loss: 0.8267 - inversion_loss: 0.7364 - root_loss: 0.9487 - key_ac 12/197 [>.............................] - ETA: 1:47 - loss: 4.4423 - key_loss: 0.7873 - degree_1_loss: 0.3256 - degree_2_loss: 0.8338 - quality_loss: 0.8228 - inversion_loss: 0.7327 - root_loss: 0.9402 - key_ac 13/197 [>.............................] - ETA: 1:46 - loss: 4.4152 - key_loss: 0.7782 - degree_1_loss: 0.3248 - degree_2_loss: 0.8306 - quality_loss: 0.8193 - inversion_loss: 0.7297 - root_loss: 0.9327 - key_ac 14/197 [=>............................] - ETA: 1:45 - loss: 4.3925 - key_loss: 0.7719 - degree_1_loss: 0.3244 - degree_2_loss: 0.8275 - quality_loss: 0.8157 - inversion_loss: 0.7273 - root_loss: 0.9256 - key_ac 15/197 [=>............................] - ETA: 1:45 - loss: 4.3684 - key_loss: 0.7652 - degree_1_loss: 0.3242 - degree_2_loss: 0.8240 - quality_loss: 0.8120 - inversion_loss: 0.7245 - root_loss: 0.9184 - key_ac 16/197 [=>............................] - ETA: 1:44 - loss: 4.3530 - key_loss: 0.7608 - degree_1_loss: 0.3242 - degree_2_loss: 0.8216 - quality_loss: 0.8095 - inversion_loss: 0.7233 - root_loss: 0.9136 - key_ac 17/197 [=>............................] - ETA: 1:43 - loss: 4.3400 - key_loss: 0.7578 - degree_1_loss: 0.3239 - degree_2_loss: 0.8194 - quality_loss: 0.8072 - inversion_loss: 0.7220 - root_loss: 0.9097 - key_ac 18/197 [=>............................] - ETA: 1:43 - loss: 4.3300 - key_loss: 0.7557 - degree_1_loss: 0.3236 - degree_2_loss: 0.8177 - quality_loss: 0.8054 - inversion_loss: 0.7208 - root_loss: 0.9068 - key_ac 19/197 [=>............................] - ETA: 1:42 - loss: 4.3189 - key_loss: 0.7532 - degree_1_loss: 0.3229 - degree_2_loss: 0.8159 - quality_loss: 0.8034 - inversion_loss: 0.7196 - root_loss: 0.9038 - key_ac 20/197 [==>...........................] - ETA: 1:42 - loss: 4.3117 - key_loss: 0.7520 - degree_1_loss: 0.3224 - degree_2_loss: 0.8148 - quality_loss: 0.8020 - inversion_loss: 0.7185 - root_loss: 0.9020 - key_ac 21/197 [==>...........................] - ETA: 1:41 - loss: 4.3051 - key_loss: 0.7511 - degree_1_loss: 0.3220 - degree_2_loss: 0.8134 - quality_loss: 0.8006 - inversion_loss: 0.7174 - root_loss: 0.9005 - key_ac 22/197 [==>...........................] - ETA: 1:40 - loss: 4.2999 - key_loss: 0.7505 - degree_1_loss: 0.3216 - degree_2_loss: 0.8123 - quality_loss: 0.7995 - inversion_loss: 0.7165 - root_loss: 0.8995 - key_ac 23/197 [==>...........................] - ETA: 1:40 - loss: 4.2955 - key_loss: 0.7500 - degree_1_loss: 0.3212 - degree_2_loss: 0.8114 - quality_loss: 0.7984 - inversion_loss: 0.7159 - root_loss: 0.8985 - key_ac 24/197 [==>...........................] - ETA: 1:39 - loss: 4.2918 - key_loss: 0.7497 - degree_1_loss: 0.3209 - degree_2_loss: 0.8105 - quality_loss: 0.7975 - inversion_loss: 0.7154 - root_loss: 0.8978 - key_ac 25/197 [==>...........................] - ETA: 1:39 - loss: 4.2909 - key_loss: 0.7503 - degree_1_loss: 0.3205 - degree_2_loss: 0.8101 - quality_loss: 0.7970 - inversion_loss: 0.7152 - root_loss: 0.8978 - key_ac 26/197 [==>...........................] - ETA: 1:39 - loss: 4.2875 - key_loss: 0.7503 - degree_1_loss: 0.3200 - degree_2_loss: 0.8093 - quality_loss: 0.7961 - inversion_loss: 0.7147 - root_loss: 0.8971 - key_ac 27/197 [===>..........................] - ETA: 1:38 - loss: 4.2848 - key_loss: 0.7506 - degree_1_loss: 0.3196 - degree_2_loss: 0.8085 - quality_loss: 0.7954 - inversion_loss: 0.7142 - root_loss: 0.8964 - key_ac 28/197 [===>..........................] - ETA: 1:37 - loss: 4.2823 - key_loss: 0.7508 - degree_1_loss: 0.3194 - degree_2_loss: 0.8079 - quality_loss: 0.7947 - inversion_loss: 0.7137 - root_loss: 0.8957 - key_ac 29/197 [===>..........................] - ETA: 1:37 - loss: 4.2792 - key_loss: 0.7507 - degree_1_loss: 0.3192 - degree_2_loss: 0.8072 - quality_loss: 0.7940 - inversion_loss: 0.7131 - root_loss: 0.8949 - key_ac 30/197 [===>..........................] - ETA: 1:36 - loss: 4.2776 - key_loss: 0.7508 - degree_1_loss: 0.3190 - degree_2_loss: 0.8069 - quality_loss: 0.7936 - inversion_loss: 0.7127 - root_loss: 0.8945 - key_ac 31/197 [===>..........................] - ETA: 1:35 - loss: 4.2759 - key_loss: 0.7507 - degree_1_loss: 0.3188 - degree_2_loss: 0.8066 - quality_loss: 0.7932 - inversion_loss: 0.7122 - root_loss: 0.8943 - key_ac 32/197 [===>..........................] - ETA: 1:35 - loss: 4.2736 - key_loss: 0.7505 - degree_1_loss: 0.3187 - degree_2_loss: 0.8063 - quality_loss: 0.7928 - inversion_loss: 0.7116 - root_loss: 0.8937 - key_ac 33/197 [====>.........................] - ETA: 1:34 - loss: 4.2710 - key_loss: 0.7501 - degree_1_loss: 0.3185 - degree_2_loss: 0.8061 - quality_loss: 0.7923 - inversion_loss: 0.7110 - root_loss: 0.8930 - key_ac 34/197 [====>.........................] - ETA: 1:34 - loss: 4.2683 - key_loss: 0.7496 - degree_1_loss: 0.3184 - degree_2_loss: 0.8058 - quality_loss: 0.7918 - inversion_loss: 0.7105 - root_loss: 0.8922 - key_ac 35/197 [====>.........................] - ETA: 1:33 - loss: 4.2651 - key_loss: 0.7490 - degree_1_loss: 0.3183 - degree_2_loss: 0.8055 - quality_loss: 0.7912 - inversion_loss: 0.7099 - root_loss: 0.8913 - key_ac 36/197 [====>.........................] - ETA: 1:33 - loss: 4.2622 - key_loss: 0.7485 - degree_1_loss: 0.3181 - degree_2_loss: 0.8051 - quality_loss: 0.7907 - inversion_loss: 0.7093 - root_loss: 0.8904 - key_ac 37/197 [====>.........................] - ETA: 1:32 - loss: 4.2586 - key_loss: 0.7479 - degree_1_loss: 0.3179 - degree_2_loss: 0.8047 - quality_loss: 0.7901 - inversion_loss: 0.7087 - root_loss: 0.8893 - key_ac 38/197 [====>.........................] - ETA: 1:32 - loss: 4.2563 - key_loss: 0.7475 - degree_1_loss: 0.3177 - degree_2_loss: 0.8045 - quality_loss: 0.7897 - inversion_loss: 0.7082 - root_loss: 0.8886 - key_ac 39/197 [====>.........................] - ETA: 1:31 - loss: 4.2542 - key_loss: 0.7472 - degree_1_loss: 0.3176 - degree_2_loss: 0.8044 - quality_loss: 0.7893 - inversion_loss: 0.7078 - root_loss: 0.8879 - key_ac 40/197 [=====>........................] - ETA: 1:31 - loss: 4.2520 - key_loss: 0.7468 - degree_1_loss: 0.3174 - degree_2_loss: 0.8044 - quality_loss: 0.7890 - inversion_loss: 0.7073 - root_loss: 0.8872 - key_ac 41/197 [=====>........................] - ETA: 1:30 - loss: 4.2504 - key_loss: 0.7465 - degree_1_loss: 0.3173 - degree_2_loss: 0.8044 - quality_loss: 0.7886 - inversion_loss: 0.7069 - root_loss: 0.8867 - key_ac 42/197 [=====>........................] - ETA: 1:29 - loss: 4.2484 - key_loss: 0.7461 - degree_1_loss: 0.3172 - degree_2_loss: 0.8043 - quality_loss: 0.7883 - inversion_loss: 0.7064 - root_loss: 0.8861 - key_ac 43/197 [=====>........................] - ETA: 1:29 - loss: 4.2464 - key_loss: 0.7456 - degree_1_loss: 0.3171 - degree_2_loss: 0.8042 - quality_loss: 0.7879 - inversion_loss: 0.7060 - root_loss: 0.8855 - key_ac 44/197 [=====>........................] - ETA: 1:28 - loss: 4.2447 - key_loss: 0.7452 - degree_1_loss: 0.3171 - degree_2_loss: 0.8042 - quality_loss: 0.7876 - inversion_loss: 0.7057 - root_loss: 0.8850 - key_ac 45/197 [=====>........................] - ETA: 1:28 - loss: 4.2425 - key_loss: 0.7446 - degree_1_loss: 0.3171 - degree_2_loss: 0.8040 - quality_loss: 0.7872 - inversion_loss: 0.7053 - root_loss: 0.8843 - key_ac 46/197 [======>.......................] - ETA: 1:27 - loss: 4.2403 - key_loss: 0.7440 - degree_1_loss: 0.3171 - degree_2_loss: 0.8039 - quality_loss: 0.7868 - inversion_loss: 0.7049 - root_loss: 0.8836 - key_ac 47/197 [======>.......................] - ETA: 1:27 - loss: 4.2378 - key_loss: 0.7433 - degree_1_loss: 0.3171 - degree_2_loss: 0.8037 - quality_loss: 0.7864 - inversion_loss: 0.7045 - root_loss: 0.8828 - key_ac 48/197 [======>.......................] - ETA: 1:26 - loss: 4.2349 - key_loss: 0.7425 - degree_1_loss: 0.3170 - degree_2_loss: 0.8034 - quality_loss: 0.7860 - inversion_loss: 0.7040 - root_loss: 0.8820 - key_ac 49/197 [======>.......................] - ETA: 1:25 - loss: 4.2319 - key_loss: 0.7416 - degree_1_loss: 0.3169 - degree_2_loss: 0.8032 - quality_loss: 0.7855 - inversion_loss: 0.7036 - root_loss: 0.8811 - key_ac 50/197 [======>.......................] - ETA: 1:25 - loss: 4.2290 - key_loss: 0.7406 - degree_1_loss: 0.3169 - degree_2_loss: 0.8029 - quality_loss: 0.7851 - inversion_loss: 0.7032 - root_loss: 0.8803 - key_ac 51/197 [======>.......................] - ETA: 1:24 - loss: 4.2261 - key_loss: 0.7397 - degree_1_loss: 0.3169 - degree_2_loss: 0.8026 - quality_loss: 0.7846 - inversion_loss: 0.7028 - root_loss: 0.8794 - key_ac 52/197 [======>.......................] - ETA: 1:24 - loss: 4.2230 - key_loss: 0.7387 - degree_1_loss: 0.3169 - degree_2_loss: 0.8023 - quality_loss: 0.7841 - inversion_loss: 0.7025 - root_loss: 0.8786 - key_ac 53/197 [=======>......................] - ETA: 1:23 - loss: 4.2201 - key_loss: 0.7377 - degree_1_loss: 0.3169 - degree_2_loss: 0.8020 - quality_loss: 0.7836 - inversion_loss: 0.7021 - root_loss: 0.8777 - key_ac 54/197 [=======>......................] - ETA: 1:22 - loss: 4.2172 - key_loss: 0.7367 - degree_1_loss: 0.3169 - degree_2_loss: 0.8017 - quality_loss: 0.7831 - inversion_loss: 0.7018 - root_loss: 0.8769 - key_ac 55/197 [=======>......................] - ETA: 1:22 - loss: 4.2144 - key_loss: 0.7358 - degree_1_loss: 0.3168 - degree_2_loss: 0.8015 - quality_loss: 0.7826 - inversion_loss: 0.7015 - root_loss: 0.8762 - key_ac 56/197 [=======>......................] - ETA: 1:21 - loss: 4.2121 - key_loss: 0.7350 - degree_1_loss: 0.3168 - degree_2_loss: 0.8013 - quality_loss: 0.7822 - inversion_loss: 0.7012 - root_loss: 0.8756 - key_ac 57/197 [=======>......................] - ETA: 1:21 - loss: 4.2099 - key_loss: 0.7342 - degree_1_loss: 0.3168 - degree_2_loss: 0.8011 - quality_loss: 0.7819 - inversion_loss: 0.7009 - root_loss: 0.8750 - key_ac 58/197 [=======>......................] - ETA: 1:20 - loss: 4.2078 - key_loss: 0.7334 - degree_1_loss: 0.3168 - degree_2_loss: 0.8010 - quality_loss: 0.7815 - inversion_loss: 0.7007 - root_loss: 0.8745 - key_ac 59/197 [=======>......................] - ETA: 1:20 - loss: 4.2057 - key_loss: 0.7326 - degree_1_loss: 0.3168 - degree_2_loss: 0.8008 - quality_loss: 0.7812 - inversion_loss: 0.7004 - root_loss: 0.8739 - key_ac 60/197 [========>.....................] - ETA: 1:19 - loss: 4.2037 - key_loss: 0.7318 - degree_1_loss: 0.3168 - degree_2_loss: 0.8006 - quality_loss: 0.7809 - inversion_loss: 0.7002 - root_loss: 0.8734 - key_ac 61/197 [========>.....................] - ETA: 1:18 - loss: 4.2014 - key_loss: 0.7309 - degree_1_loss: 0.3168 - degree_2_loss: 0.8004 - quality_loss: 0.7805 - inversion_loss: 0.6999 - root_loss: 0.8728 - key_ac 62/197 [========>.....................] - ETA: 1:18 - loss: 4.1990 - key_loss: 0.7301 - degree_1_loss: 0.3168 - degree_2_loss: 0.8002 - quality_loss: 0.7802 - inversion_loss: 0.6996 - root_loss: 0.8722 - key_ac 63/197 [========>.....................] - ETA: 1:17 - loss: 4.1974 - key_loss: 0.7295 - degree_1_loss: 0.3168 - degree_2_loss: 0.8000 - quality_loss: 0.7799 - inversion_loss: 0.6994 - root_loss: 0.8718 - key_ac 64/197 [========>.....................] - ETA: 1:17 - loss: 4.1956 - key_loss: 0.7289 - degree_1_loss: 0.3168 - degree_2_loss: 0.7998 - quality_loss: 0.7796 - inversion_loss: 0.6991 - root_loss: 0.8713 - key_ac 65/197 [========>.....................] - ETA: 1:16 - loss: 4.1937 - key_loss: 0.7283 - degree_1_loss: 0.3168 - degree_2_loss: 0.7996 - quality_loss: 0.7793 - inversion_loss: 0.6989 - root_loss: 0.8708 - key_ac 66/197 [=========>....................] - ETA: 1:16 - loss: 4.1923 - key_loss: 0.7278 - degree_1_loss: 0.3168 - degree_2_loss: 0.7995 - quality_loss: 0.7791 - inversion_loss: 0.6987 - root_loss: 0.8704 - key_ac 67/197 [=========>....................] - ETA: 1:15 - loss: 4.1907 - key_loss: 0.7273 - degree_1_loss: 0.3168 - degree_2_loss: 0.7993 - quality_loss: 0.7789 - inversion_loss: 0.6985 - root_loss: 0.8700 - key_ac 68/197 [=========>....................] - ETA: 1:14 - loss: 4.1889 - key_loss: 0.7266 - degree_1_loss: 0.3167 - degree_2_loss: 0.7992 - quality_loss: 0.7786 - inversion_loss: 0.6983 - root_loss: 0.8695 - key_ac 69/197 [=========>....................] - ETA: 1:14 - loss: 4.1871 - key_loss: 0.7260 - degree_1_loss: 0.3167 - degree_2_loss: 0.7990 - quality_loss: 0.7783 - inversion_loss: 0.6981 - root_loss: 0.8691 - key_ac 70/197 [=========>....................] - ETA: 1:13 - loss: 4.1854 - key_loss: 0.7253 - degree_1_loss: 0.3166 - degree_2_loss: 0.7988 - quality_loss: 0.7781 - inversion_loss: 0.6979 - root_loss: 0.8686 - key_ac 71/197 [=========>....................] - ETA: 1:13 - loss: 4.1837 - key_loss: 0.7247 - degree_1_loss: 0.3166 - degree_2_loss: 0.7987 - quality_loss: 0.7778 - inversion_loss: 0.6976 - root_loss: 0.8682 - key_ac 72/197 [=========>....................] - ETA: 1:12 - loss: 4.1823 - key_loss: 0.7242 - degree_1_loss: 0.3166 - degree_2_loss: 0.7986 - quality_loss: 0.7777 - inversion_loss: 0.6975 - root_loss: 0.8678 - key_ac 73/197 [==========>...................] - ETA: 1:12 - loss: 4.1810 - key_loss: 0.7238 - degree_1_loss: 0.3165 - degree_2_loss: 0.7984 - quality_loss: 0.7775 - inversion_loss: 0.6973 - root_loss: 0.8675 - key_ac 74/197 [==========>...................] - ETA: 1:11 - loss: 4.1796 - key_loss: 0.7233 - degree_1_loss: 0.3165 - degree_2_loss: 0.7983 - quality_loss: 0.7773 - inversion_loss: 0.6971 - root_loss: 0.8671 - key_ac 75/197 [==========>...................] - ETA: 1:10 - loss: 4.1783 - key_loss: 0.7228 - degree_1_loss: 0.3165 - degree_2_loss: 0.7982 - quality_loss: 0.7771 - inversion_loss: 0.6970 - root_loss: 0.8667 - key_ac 76/197 [==========>...................] - ETA: 1:10 - loss: 4.1769 - key_loss: 0.7223 - degree_1_loss: 0.3165 - degree_2_loss: 0.7980 - quality_loss: 0.7769 - inversion_loss: 0.6968 - root_loss: 0.8664 - key_ac 77/197 [==========>...................] - ETA: 1:09 - loss: 4.1754 - key_loss: 0.7217 - degree_1_loss: 0.3165 - degree_2_loss: 0.7979 - quality_loss: 0.7767 - inversion_loss: 0.6966 - root_loss: 0.8660 - key_ac 78/197 [==========>...................] - ETA: 1:09 - loss: 4.1742 - key_loss: 0.7213 - degree_1_loss: 0.3165 - degree_2_loss: 0.7978 - quality_loss: 0.7766 - inversion_loss: 0.6964 - root_loss: 0.8657 - key_ac 79/197 [===========>..................] - ETA: 1:08 - loss: 4.1730 - key_loss: 0.7209 - degree_1_loss: 0.3164 - degree_2_loss: 0.7977 - quality_loss: 0.7764 - inversion_loss: 0.6963 - root_loss: 0.8653 - key_ac 80/197 [===========>..................] - ETA: 1:07 - loss: 4.1717 - key_loss: 0.7205 - degree_1_loss: 0.3164 - degree_2_loss: 0.7975 - quality_loss: 0.7762 - inversion_loss: 0.6961 - root_loss: 0.8649 - key_ac 81/197 [===========>..................] - ETA: 1:07 - loss: 4.1703 - key_loss: 0.7200 - degree_1_loss: 0.3164 - degree_2_loss: 0.7974 - quality_loss: 0.7760 - inversion_loss: 0.6959 - root_loss: 0.8646 - key_ac 82/197 [===========>..................] - ETA: 1:06 - loss: 4.1688 - key_loss: 0.7196 - degree_1_loss: 0.3163 - degree_2_loss: 0.7972 - quality_loss: 0.7758 - inversion_loss: 0.6957 - root_loss: 0.8642 - key_ac 83/197 [===========>..................] - ETA: 1:06 - loss: 4.1676 - key_loss: 0.7191 - degree_1_loss: 0.3163 - degree_2_loss: 0.7971 - quality_loss: 0.7756 - inversion_loss: 0.6955 - root_loss: 0.8639 - key_ac 84/197 [===========>..................] - ETA: 1:05 - loss: 4.1663 - key_loss: 0.7187 - degree_1_loss: 0.3163 - degree_2_loss: 0.7970 - quality_loss: 0.7755 - inversion_loss: 0.6953 - root_loss: 0.8636 - key_ac 85/197 [===========>..................] - ETA: 1:05 - loss: 4.1652 - key_loss: 0.7183 - degree_1_loss: 0.3163 - degree_2_loss: 0.7969 - quality_loss: 0.7753 - inversion_loss: 0.6951 - root_loss: 0.8632 - key_ac 86/197 [============>.................] - ETA: 1:04 - loss: 4.1641 - key_loss: 0.7180 - degree_1_loss: 0.3163 - degree_2_loss: 0.7968 - quality_loss: 0.7752 - inversion_loss: 0.6950 - root_loss: 0.8630 - key_ac 87/197 [============>.................] - ETA: 1:03 - loss: 4.1631 - key_loss: 0.7176 - degree_1_loss: 0.3162 - degree_2_loss: 0.7967 - quality_loss: 0.7750 - inversion_loss: 0.6948 - root_loss: 0.8627 - key_ac 88/197 [============>.................] - ETA: 1:03 - loss: 4.1622 - key_loss: 0.7173 - degree_1_loss: 0.3162 - degree_2_loss: 0.7967 - quality_loss: 0.7749 - inversion_loss: 0.6946 - root_loss: 0.8624 - key_ac 89/197 [============>.................] - ETA: 1:02 - loss: 4.1611 - key_loss: 0.7169 - degree_1_loss: 0.3162 - degree_2_loss: 0.7966 - quality_loss: 0.7748 - inversion_loss: 0.6945 - root_loss: 0.8621 - key_ac 90/197 [============>.................] - ETA: 1:02 - loss: 4.1600 - key_loss: 0.7166 - degree_1_loss: 0.3162 - degree_2_loss: 0.7965 - quality_loss: 0.7747 - inversion_loss: 0.6943 - root_loss: 0.8618 - key_ac 91/197 [============>.................] - ETA: 1:01 - loss: 4.1590 - key_loss: 0.7162 - degree_1_loss: 0.3162 - degree_2_loss: 0.7964 - quality_loss: 0.7745 - inversion_loss: 0.6941 - root_loss: 0.8615 - key_ac 92/197 [=============>................] - ETA: 1:01 - loss: 4.1581 - key_loss: 0.7159 - degree_1_loss: 0.3162 - degree_2_loss: 0.7963 - quality_loss: 0.7744 - inversion_loss: 0.6940 - root_loss: 0.8613 - key_ac 93/197 [=============>................] - ETA: 1:00 - loss: 4.1571 - key_loss: 0.7156 - degree_1_loss: 0.3162 - degree_2_loss: 0.7962 - quality_loss: 0.7743 - inversion_loss: 0.6939 - root_loss: 0.8610 - key_ac 94/197 [=============>................] - ETA: 59s - loss: 4.1562 - key_loss: 0.7153 - degree_1_loss: 0.3162 - degree_2_loss: 0.7961 - quality_loss: 0.7742 - inversion_loss: 0.6937 - root_loss: 0.8607 - key_acc 95/197 [=============>................] - ETA: 59s - loss: 4.1554 - key_loss: 0.7150 - degree_1_loss: 0.3162 - degree_2_loss: 0.7961 - quality_loss: 0.7741 - inversion_loss: 0.6936 - root_loss: 0.8604 - key_acc 96/197 [=============>................] - ETA: 58s - loss: 4.1546 - key_loss: 0.7147 - degree_1_loss: 0.3162 - degree_2_loss: 0.7960 - quality_loss: 0.7740 - inversion_loss: 0.6935 - root_loss: 0.8602 - key_acc 97/197 [=============>................] - ETA: 58s - loss: 4.1539 - key_loss: 0.7145 - degree_1_loss: 0.3162 - degree_2_loss: 0.7959 - quality_loss: 0.7739 - inversion_loss: 0.6934 - root_loss: 0.8600 - key_acc 98/197 [=============>................] - ETA: 57s - loss: 4.1533 - key_loss: 0.7142 - degree_1_loss: 0.3163 - degree_2_loss: 0.7959 - quality_loss: 0.7739 - inversion_loss: 0.6933 - root_loss: 0.8597 - key_acc 99/197 [==============>...............] - ETA: 56s - loss: 4.1526 - key_loss: 0.7140 - degree_1_loss: 0.3163 - degree_2_loss: 0.7958 - quality_loss: 0.7738 - inversion_loss: 0.6932 - root_loss: 0.8595 - key_acc100/197 [==============>...............] - ETA: 56s - loss: 4.1522 - key_loss: 0.7138 - degree_1_loss: 0.3163 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6932 - root_loss: 0.8593 - key_acc101/197 [==============>...............] - ETA: 55s - loss: 4.1519 - key_loss: 0.7137 - degree_1_loss: 0.3164 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6931 - root_loss: 0.8592 - key_acc102/197 [==============>...............] - ETA: 55s - loss: 4.1516 - key_loss: 0.7136 - degree_1_loss: 0.3165 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6931 - root_loss: 0.8590 - key_acc103/197 [==============>...............] - ETA: 54s - loss: 4.1514 - key_loss: 0.7135 - degree_1_loss: 0.3165 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6930 - root_loss: 0.8589 - key_acc104/197 [==============>...............] - ETA: 54s - loss: 4.1512 - key_loss: 0.7133 - degree_1_loss: 0.3166 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6930 - root_loss: 0.8588 - key_acc105/197 [==============>...............] - ETA: 53s - loss: 4.1510 - key_loss: 0.7132 - degree_1_loss: 0.3166 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6929 - root_loss: 0.8587 - key_acc106/197 [===============>..............] - ETA: 52s - loss: 4.1507 - key_loss: 0.7131 - degree_1_loss: 0.3167 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6929 - root_loss: 0.8586 - key_acc107/197 [===============>..............] - ETA: 52s - loss: 4.1505 - key_loss: 0.7130 - degree_1_loss: 0.3167 - degree_2_loss: 0.7958 - quality_loss: 0.7737 - inversion_loss: 0.6928 - root_loss: 0.8584 - key_acc108/197 [===============>..............] - ETA: 51s - loss: 4.1504 - key_loss: 0.7129 - degree_1_loss: 0.3167 - degree_2_loss: 0.7959 - quality_loss: 0.7738 - inversion_loss: 0.6928 - root_loss: 0.8583 - key_acc109/197 [===============>..............] - ETA: 51s - loss: 4.1502 - key_loss: 0.7128 - degree_1_loss: 0.3168 - degree_2_loss: 0.7959 - quality_loss: 0.7738 - inversion_loss: 0.6928 - root_loss: 0.8582 - key_acc110/197 [===============>..............] - ETA: 50s - loss: 4.1502 - key_loss: 0.7127 - degree_1_loss: 0.3168 - degree_2_loss: 0.7959 - quality_loss: 0.7738 - inversion_loss: 0.6928 - root_loss: 0.8582 - key_acc111/197 [===============>..............] - ETA: 49s - loss: 4.1501 - key_loss: 0.7126 - degree_1_loss: 0.3168 - degree_2_loss: 0.7960 - quality_loss: 0.7738 - inversion_loss: 0.6928 - root_loss: 0.8581 - key_acc112/197 [================>.............] - ETA: 49s - loss: 4.1500 - key_loss: 0.7125 - degree_1_loss: 0.3169 - degree_2_loss: 0.7960 - quality_loss: 0.7738 - inversion_loss: 0.6927 - root_loss: 0.8580 - key_acc113/197 [================>.............] - ETA: 48s - loss: 4.1499 - key_loss: 0.7124 - degree_1_loss: 0.3169 - degree_2_loss: 0.7960 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8579 - key_acc114/197 [================>.............] - ETA: 48s - loss: 4.1499 - key_loss: 0.7124 - degree_1_loss: 0.3170 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8579 - key_acc115/197 [================>.............] - ETA: 47s - loss: 4.1498 - key_loss: 0.7123 - degree_1_loss: 0.3170 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8578 - key_acc116/197 [================>.............] - ETA: 47s - loss: 4.1497 - key_loss: 0.7123 - degree_1_loss: 0.3170 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8577 - key_acc117/197 [================>.............] - ETA: 46s - loss: 4.1496 - key_loss: 0.7122 - degree_1_loss: 0.3171 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8576 - key_acc118/197 [================>.............] - ETA: 45s - loss: 4.1497 - key_loss: 0.7122 - degree_1_loss: 0.3171 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8576 - key_acc119/197 [=================>............] - ETA: 45s - loss: 4.1497 - key_loss: 0.7122 - degree_1_loss: 0.3172 - degree_2_loss: 0.7962 - quality_loss: 0.7739 - inversion_loss: 0.6927 - root_loss: 0.8576 - key_acc120/197 [=================>............] - ETA: 44s - loss: 4.1497 - key_loss: 0.7121 - degree_1_loss: 0.3172 - degree_2_loss: 0.7962 - quality_loss: 0.7740 - inversion_loss: 0.6927 - root_loss: 0.8575 - key_acc121/197 [=================>............] - ETA: 44s - loss: 4.1496 - key_loss: 0.7121 - degree_1_loss: 0.3172 - degree_2_loss: 0.7962 - quality_loss: 0.7740 - inversion_loss: 0.6927 - root_loss: 0.8575 - key_acc122/197 [=================>............] - ETA: 43s - loss: 4.1495 - key_loss: 0.7120 - degree_1_loss: 0.3173 - degree_2_loss: 0.7962 - quality_loss: 0.7740 - inversion_loss: 0.6926 - root_loss: 0.8574 - key_acc123/197 [=================>............] - ETA: 42s - loss: 4.1494 - key_loss: 0.7119 - degree_1_loss: 0.3173 - degree_2_loss: 0.7963 - quality_loss: 0.7740 - inversion_loss: 0.6926 - root_loss: 0.8574 - key_acc124/197 [=================>............] - ETA: 42s - loss: 4.1493 - key_loss: 0.7118 - degree_1_loss: 0.3173 - degree_2_loss: 0.7963 - quality_loss: 0.7740 - inversion_loss: 0.6926 - root_loss: 0.8573 - key_acc125/197 [==================>...........] - ETA: 41s - loss: 4.1492 - key_loss: 0.7117 - degree_1_loss: 0.3174 - degree_2_loss: 0.7963 - quality_loss: 0.7740 - inversion_loss: 0.6926 - root_loss: 0.8572 - key_acc126/197 [==================>...........] - ETA: 41s - loss: 4.1490 - key_loss: 0.7116 - degree_1_loss: 0.3174 - degree_2_loss: 0.7963 - quality_loss: 0.7740 - inversion_loss: 0.6926 - root_loss: 0.8571 - key_acc127/197 [==================>...........] - ETA: 40s - loss: 4.1489 - key_loss: 0.7115 - degree_1_loss: 0.3175 - degree_2_loss: 0.7963 - quality_loss: 0.7740 - inversion_loss: 0.6926 - root_loss: 0.8571 - key_acc128/197 [==================>...........] - ETA: 40s - loss: 4.1489 - key_loss: 0.7114 - degree_1_loss: 0.3175 - degree_2_loss: 0.7963 - quality_loss: 0.7741 - inversion_loss: 0.6926 - root_loss: 0.8570 - key_acc129/197 [==================>...........] - ETA: 39s - loss: 4.1488 - key_loss: 0.7113 - degree_1_loss: 0.3175 - degree_2_loss: 0.7964 - quality_loss: 0.7741 - inversion_loss: 0.6926 - root_loss: 0.8570 - key_acc130/197 [==================>...........] - ETA: 38s - loss: 4.1487 - key_loss: 0.7112 - degree_1_loss: 0.3176 - degree_2_loss: 0.7964 - quality_loss: 0.7741 - inversion_loss: 0.6926 - root_loss: 0.8569 - key_acc131/197 [==================>...........] - ETA: 38s - loss: 4.1485 - key_loss: 0.7110 - degree_1_loss: 0.3176 - degree_2_loss: 0.7964 - quality_loss: 0.7741 - inversion_loss: 0.6926 - root_loss: 0.8568 - key_acc132/197 [===================>..........] - ETA: 37s - loss: 4.1483 - key_loss: 0.7109 - degree_1_loss: 0.3176 - degree_2_loss: 0.7964 - quality_loss: 0.7741 - inversion_loss: 0.6925 - root_loss: 0.8567 - key_acc133/197 [===================>..........] - ETA: 37s - loss: 4.1481 - key_loss: 0.7108 - degree_1_loss: 0.3177 - degree_2_loss: 0.7964 - quality_loss: 0.7741 - inversion_loss: 0.6925 - root_loss: 0.8566 - key_acc134/197 [===================>..........] - ETA: 36s - loss: 4.1479 - key_loss: 0.7106 - degree_1_loss: 0.3177 - degree_2_loss: 0.7964 - quality_loss: 0.7741 - inversion_loss: 0.6925 - root_loss: 0.8565 - key_acc135/197 [===================>..........] - ETA: 35s - loss: 4.1476 - key_loss: 0.7105 - degree_1_loss: 0.3177 - degree_2_loss: 0.7964 - quality_loss: 0.7742 - inversion_loss: 0.6925 - root_loss: 0.8564 - key_acc136/197 [===================>..........] - ETA: 35s - loss: 4.1472 - key_loss: 0.7103 - degree_1_loss: 0.3178 - degree_2_loss: 0.7963 - quality_loss: 0.7741 - inversion_loss: 0.6924 - root_loss: 0.8563 - key_acc137/197 [===================>..........] - ETA: 34s - loss: 4.1469 - key_loss: 0.7101 - degree_1_loss: 0.3178 - degree_2_loss: 0.7963 - quality_loss: 0.7741 - inversion_loss: 0.6924 - root_loss: 0.8561 - key_acc138/197 [====================>.........] - ETA: 34s - loss: 4.1466 - key_loss: 0.7100 - degree_1_loss: 0.3178 - degree_2_loss: 0.7963 - quality_loss: 0.7741 - inversion_loss: 0.6923 - root_loss: 0.8560 - key_acc139/197 [====================>.........] - ETA: 33s - loss: 4.1463 - key_loss: 0.7098 - degree_1_loss: 0.3178 - degree_2_loss: 0.7963 - quality_loss: 0.7741 - inversion_loss: 0.6923 - root_loss: 0.8559 - key_acc140/197 [====================>.........] - ETA: 33s - loss: 4.1459 - key_loss: 0.7097 - degree_1_loss: 0.3178 - degree_2_loss: 0.7963 - quality_loss: 0.7741 - inversion_loss: 0.6923 - root_loss: 0.8558 - key_acc141/197 [====================>.........] - ETA: 32s - loss: 4.1456 - key_loss: 0.7095 - degree_1_loss: 0.3178 - degree_2_loss: 0.7962 - quality_loss: 0.7741 - inversion_loss: 0.6922 - root_loss: 0.8557 - key_acc142/197 [====================>.........] - ETA: 31s - loss: 4.1452 - key_loss: 0.7093 - degree_1_loss: 0.3179 - degree_2_loss: 0.7962 - quality_loss: 0.7740 - inversion_loss: 0.6922 - root_loss: 0.8556 - key_acc143/197 [====================>.........] - ETA: 31s - loss: 4.1449 - key_loss: 0.7092 - degree_1_loss: 0.3179 - degree_2_loss: 0.7962 - quality_loss: 0.7740 - inversion_loss: 0.6921 - root_loss: 0.8555 - key_acc144/197 [====================>.........] - ETA: 30s - loss: 4.1445 - key_loss: 0.7090 - degree_1_loss: 0.3179 - degree_2_loss: 0.7961 - quality_loss: 0.7740 - inversion_loss: 0.6921 - root_loss: 0.8553 - key_acc145/197 [=====================>........] - ETA: 30s - loss: 4.1441 - key_loss: 0.7088 - degree_1_loss: 0.3180 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6920 - root_loss: 0.8552 - key_acc146/197 [=====================>........] - ETA: 29s - loss: 4.1437 - key_loss: 0.7087 - degree_1_loss: 0.3180 - degree_2_loss: 0.7961 - quality_loss: 0.7739 - inversion_loss: 0.6920 - root_loss: 0.8551 - key_acc147/197 [=====================>........] - ETA: 29s - loss: 4.1434 - key_loss: 0.7085 - degree_1_loss: 0.3180 - degree_2_loss: 0.7960 - quality_loss: 0.7739 - inversion_loss: 0.6919 - root_loss: 0.8550 - key_acc148/197 [=====================>........] - ETA: 28s - loss: 4.1429 - key_loss: 0.7083 - degree_1_loss: 0.3181 - degree_2_loss: 0.7960 - quality_loss: 0.7738 - inversion_loss: 0.6919 - root_loss: 0.8549 - key_acc149/197 [=====================>........] - ETA: 27s - loss: 4.1425 - key_loss: 0.7081 - degree_1_loss: 0.3181 - degree_2_loss: 0.7960 - quality_loss: 0.7738 - inversion_loss: 0.6918 - root_loss: 0.8547 - key_acc150/197 [=====================>........] - ETA: 27s - loss: 4.1421 - key_loss: 0.7079 - degree_1_loss: 0.3181 - degree_2_loss: 0.7959 - quality_loss: 0.7737 - inversion_loss: 0.6918 - root_loss: 0.8546 - key_acc151/197 [=====================>........] - ETA: 26s - loss: 4.1416 - key_loss: 0.7078 - degree_1_loss: 0.3181 - degree_2_loss: 0.7959 - quality_loss: 0.7737 - inversion_loss: 0.6917 - root_loss: 0.8545 - key_acc152/197 [======================>.......] - ETA: 26s - loss: 4.1412 - key_loss: 0.7076 - degree_1_loss: 0.3182 - degree_2_loss: 0.7958 - quality_loss: 0.7736 - inversion_loss: 0.6917 - root_loss: 0.8543 - key_acc153/197 [======================>.......] - ETA: 25s - loss: 4.1409 - key_loss: 0.7074 - degree_1_loss: 0.3182 - degree_2_loss: 0.7958 - quality_loss: 0.7736 - inversion_loss: 0.6916 - root_loss: 0.8542 - key_acc154/197 [======================>.......] - ETA: 24s - loss: 4.1406 - key_loss: 0.7073 - degree_1_loss: 0.3182 - degree_2_loss: 0.7957 - quality_loss: 0.7736 - inversion_loss: 0.6916 - root_loss: 0.8541 - key_acc155/197 [======================>.......] - ETA: 24s - loss: 4.1402 - key_loss: 0.7072 - degree_1_loss: 0.3183 - degree_2_loss: 0.7957 - quality_loss: 0.7735 - inversion_loss: 0.6915 - root_loss: 0.8540 - key_acc156/197 [======================>.......] - ETA: 23s - loss: 4.1399 - key_loss: 0.7071 - degree_1_loss: 0.3183 - degree_2_loss: 0.7956 - quality_loss: 0.7735 - inversion_loss: 0.6915 - root_loss: 0.8539 - key_acc157/197 [======================>.......] - ETA: 23s - loss: 4.1396 - key_loss: 0.7070 - degree_1_loss: 0.3184 - degree_2_loss: 0.7956 - quality_loss: 0.7734 - inversion_loss: 0.6914 - root_loss: 0.8538 - key_acc158/197 [=======================>......] - ETA: 22s - loss: 4.1392 - key_loss: 0.7069 - degree_1_loss: 0.3184 - degree_2_loss: 0.7955 - quality_loss: 0.7734 - inversion_loss: 0.6914 - root_loss: 0.8537 - key_acc159/197 [=======================>......] - ETA: 22s - loss: 4.1388 - key_loss: 0.7067 - degree_1_loss: 0.3184 - degree_2_loss: 0.7955 - quality_loss: 0.7733 - inversion_loss: 0.6913 - root_loss: 0.8535 - key_acc160/197 [=======================>......] - ETA: 21s - loss: 4.1385 - key_loss: 0.7066 - degree_1_loss: 0.3185 - degree_2_loss: 0.7954 - quality_loss: 0.7733 - inversion_loss: 0.6913 - root_loss: 0.8534 - key_acc161/197 [=======================>......] - ETA: 20s - loss: 4.1382 - key_loss: 0.7065 - degree_1_loss: 0.3185 - degree_2_loss: 0.7954 - quality_loss: 0.7732 - inversion_loss: 0.6912 - root_loss: 0.8533 - key_acc162/197 [=======================>......] - ETA: 20s - loss: 4.1380 - key_loss: 0.7064 - degree_1_loss: 0.3185 - degree_2_loss: 0.7954 - quality_loss: 0.7732 - inversion_loss: 0.6912 - root_loss: 0.8532 - key_acc163/197 [=======================>......] - ETA: 19s - loss: 4.1377 - key_loss: 0.7063 - degree_1_loss: 0.3186 - degree_2_loss: 0.7953 - quality_loss: 0.7731 - inversion_loss: 0.6912 - root_loss: 0.8531 - key_acc164/197 [=======================>......] - ETA: 19s - loss: 4.1374 - key_loss: 0.7062 - degree_1_loss: 0.3186 - degree_2_loss: 0.7953 - quality_loss: 0.7731 - inversion_loss: 0.6911 - root_loss: 0.8530 - key_acc165/197 [========================>.....] - ETA: 18s - loss: 4.1371 - key_loss: 0.7061 - degree_1_loss: 0.3187 - degree_2_loss: 0.7952 - quality_loss: 0.7730 - inversion_loss: 0.6911 - root_loss: 0.8529 - key_acc166/197 [========================>.....] - ETA: 18s - loss: 4.1368 - key_loss: 0.7060 - degree_1_loss: 0.3187 - degree_2_loss: 0.7952 - quality_loss: 0.7730 - inversion_loss: 0.6911 - root_loss: 0.8528 - key_acc167/197 [========================>.....] - ETA: 17s - loss: 4.1365 - key_loss: 0.7059 - degree_1_loss: 0.3187 - degree_2_loss: 0.7952 - quality_loss: 0.7730 - inversion_loss: 0.6910 - root_loss: 0.8527 - key_acc168/197 [========================>.....] - ETA: 16s - loss: 4.1363 - key_loss: 0.7058 - degree_1_loss: 0.3188 - degree_2_loss: 0.7952 - quality_loss: 0.7729 - inversion_loss: 0.6910 - root_loss: 0.8526 - key_acc169/197 [========================>.....] - ETA: 16s - loss: 4.1360 - key_loss: 0.7057 - degree_1_loss: 0.3188 - degree_2_loss: 0.7951 - quality_loss: 0.7729 - inversion_loss: 0.6910 - root_loss: 0.8525 - key_acc170/197 [========================>.....] - ETA: 15s - loss: 4.1357 - key_loss: 0.7056 - degree_1_loss: 0.3188 - degree_2_loss: 0.7951 - quality_loss: 0.7728 - inversion_loss: 0.6909 - root_loss: 0.8524 - key_acc171/197 [=========================>....] - ETA: 15s - loss: 4.1354 - key_loss: 0.7054 - degree_1_loss: 0.3189 - degree_2_loss: 0.7951 - quality_loss: 0.7728 - inversion_loss: 0.6909 - root_loss: 0.8523 - key_acc172/197 [=========================>....] - ETA: 14s - loss: 4.1352 - key_loss: 0.7053 - degree_1_loss: 0.3189 - degree_2_loss: 0.7951 - quality_loss: 0.7728 - inversion_loss: 0.6909 - root_loss: 0.8522 - key_acc173/197 [=========================>....] - ETA: 13s - loss: 4.1350 - key_loss: 0.7053 - degree_1_loss: 0.3189 - degree_2_loss: 0.7951 - quality_loss: 0.7728 - inversion_loss: 0.6908 - root_loss: 0.8522 - key_acc174/197 [=========================>....] - ETA: 13s - loss: 4.1348 - key_loss: 0.7052 - degree_1_loss: 0.3189 - degree_2_loss: 0.7951 - quality_loss: 0.7727 - inversion_loss: 0.6908 - root_loss: 0.8521 - key_acc175/197 [=========================>....] - ETA: 12s - loss: 4.1346 - key_loss: 0.7051 - degree_1_loss: 0.3190 - degree_2_loss: 0.7950 - quality_loss: 0.7727 - inversion_loss: 0.6908 - root_loss: 0.8520 - key_acc176/197 [=========================>....] - ETA: 12s - loss: 4.1344 - key_loss: 0.7050 - degree_1_loss: 0.3190 - degree_2_loss: 0.7950 - quality_loss: 0.7727 - inversion_loss: 0.6907 - root_loss: 0.8519 - key_acc177/197 [=========================>....] - ETA: 11s - loss: 4.1342 - key_loss: 0.7049 - degree_1_loss: 0.3190 - degree_2_loss: 0.7950 - quality_loss: 0.7727 - inversion_loss: 0.6907 - root_loss: 0.8518 - key_acc178/197 [==========================>...] - ETA: 11s - loss: 4.1340 - key_loss: 0.7048 - degree_1_loss: 0.3191 - degree_2_loss: 0.7950 - quality_loss: 0.7726 - inversion_loss: 0.6907 - root_loss: 0.8518 - key_acc179/197 [==========================>...] - ETA: 10s - loss: 4.1338 - key_loss: 0.7048 - degree_1_loss: 0.3191 - degree_2_loss: 0.7950 - quality_loss: 0.7726 - inversion_loss: 0.6906 - root_loss: 0.8517 - key_acc180/197 [==========================>...] - ETA: 9s - loss: 4.1337 - key_loss: 0.7047 - degree_1_loss: 0.3191 - degree_2_loss: 0.7950 - quality_loss: 0.7726 - inversion_loss: 0.6906 - root_loss: 0.8516 - key_accu181/197 [==========================>...] - ETA: 9s - loss: 4.1335 - key_loss: 0.7046 - degree_1_loss: 0.3192 - degree_2_loss: 0.7950 - quality_loss: 0.7726 - inversion_loss: 0.6906 - root_loss: 0.8515 - key_accu182/197 [==========================>...] - ETA: 8s - loss: 4.1333 - key_loss: 0.7045 - degree_1_loss: 0.3192 - degree_2_loss: 0.7950 - quality_loss: 0.7726 - inversion_loss: 0.6906 - root_loss: 0.8514 - key_accu183/197 [==========================>...] - ETA: 8s - loss: 4.1331 - key_loss: 0.7044 - degree_1_loss: 0.3192 - degree_2_loss: 0.7950 - quality_loss: 0.7725 - inversion_loss: 0.6905 - root_loss: 0.8514 - key_accu184/197 [===========================>..] - ETA: 7s - loss: 4.1329 - key_loss: 0.7044 - degree_1_loss: 0.3193 - degree_2_loss: 0.7950 - quality_loss: 0.7725 - inversion_loss: 0.6905 - root_loss: 0.8513 - key_accu185/197 [===========================>..] - ETA: 6s - loss: 4.1328 - key_loss: 0.7043 - degree_1_loss: 0.3193 - degree_2_loss: 0.7950 - quality_loss: 0.7725 - inversion_loss: 0.6905 - root_loss: 0.8512 - key_accu186/197 [===========================>..] - ETA: 6s - loss: 4.1327 - key_loss: 0.7043 - degree_1_loss: 0.3193 - degree_2_loss: 0.7950 - quality_loss: 0.7725 - inversion_loss: 0.6905 - root_loss: 0.8512 - key_accu187/197 [===========================>..] - ETA: 5s - loss: 4.1325 - key_loss: 0.7042 - degree_1_loss: 0.3194 - degree_2_loss: 0.7950 - quality_loss: 0.7725 - inversion_loss: 0.6905 - root_loss: 0.8511 - key_accu188/197 [===========================>..] - ETA: 5s - loss: 4.1323 - key_loss: 0.7041 - degree_1_loss: 0.3194 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6904 - root_loss: 0.8510 - key_accu189/197 [===========================>..] - ETA: 4s - loss: 4.1322 - key_loss: 0.7041 - degree_1_loss: 0.3194 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6904 - root_loss: 0.8510 - key_accu190/197 [===========================>..] - ETA: 4s - loss: 4.1321 - key_loss: 0.7040 - degree_1_loss: 0.3195 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6904 - root_loss: 0.8509 - key_accu191/197 [============================>.] - ETA: 3s - loss: 4.1320 - key_loss: 0.7040 - degree_1_loss: 0.3195 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6904 - root_loss: 0.8508 - key_accu192/197 [============================>.] - ETA: 2s - loss: 4.1319 - key_loss: 0.7039 - degree_1_loss: 0.3195 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6903 - root_loss: 0.8508 - key_accu193/197 [============================>.] - ETA: 2s - loss: 4.1318 - key_loss: 0.7039 - degree_1_loss: 0.3195 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6903 - root_loss: 0.8507 - key_accu194/197 [============================>.] - ETA: 1s - loss: 4.1317 - key_loss: 0.7039 - degree_1_loss: 0.3196 - degree_2_loss: 0.7949 - quality_loss: 0.7724 - inversion_loss: 0.6903 - root_loss: 0.8507 - key_accu195/197 [============================>.] - ETA: 1s - loss: 4.1315 - key_loss: 0.7038 - degree_1_loss: 0.3196 - degree_2_loss: 0.7949 - quality_loss: 0.7723 - inversion_loss: 0.6903 - root_loss: 0.8506 - key_accu196/197 [============================>.] - ETA: 0s - loss: 4.1313 - key_loss: 0.7037 - degree_1_loss: 0.3196 - degree_2_loss: 0.7949 - quality_loss: 0.7723 - inversion_loss: 0.6902 - root_loss: 0.8505 - key_accu197/197 [==============================] - ETA: 0s - loss: 4.1311 - key_loss: 0.7037 - degree_1_loss: 0.3197 - degree_2_loss: 0.7949 - quality_loss: 0.7723 - inversion_loss: 0.6902 - root_loss: 0.8504 - key_accu197/197 [==============================] - 115s 583ms/step - loss: 4.1310 - key_loss: 0.7036 - degree_1_loss: 0.3197 - degree_2_loss: 0.7949 - quality_loss: 0.7723 - inversion_loss: 0.6902 - root_loss: 0.8504 - key_accuracy: 0.8086 - degree_1_accuracy: 0.9154 - degree_2_accuracy: 0.7294 - quality_accuracy: 0.7351 - inversion_accuracy: 0.7224 - root_accuracy: 0.7606 - val_loss: 4.3504 - val_key_loss: 0.7694 - val_degree_1_loss: 0.3549 - val_degree_2_loss: 0.8956 - val_quality_loss: 0.8011 - val_inversion_loss: 0.8197 - val_root_loss: 0.7097 - val_key_accuracy: 0.7433 - val_degree_1_accuracy: 0.9082 - val_degree_2_accuracy: 0.6919 - val_quality_accuracy: 0.7372 - val_inversion_accuracy: 0.6658 - val_root_accuracy: 0.7895 loss: 4.1310 key_loss: 0.7036 degree_1_loss: 0.3197 degree_2_loss: 0.7949 quality_loss: 0.7723 inversion_loss: 0.6902 root_loss: 0.8504 key_accuracy: 0.8086 degree_1_accuracy: 0.9154 degree_2_accuracy: 0.7294 quality_accuracy: 0.7351 inversion_accuracy: 0.7224 root_accuracy: 0.7606 val_loss: 4.3504 val_key_loss: 0.7694 val_degree_1_loss: 0.3549 val_degree_2_loss: 0.8956 val_quality_loss: 0.8011 val_inversion_loss: 0.8197 val_root_loss: 0.7097 val_key_accuracy: 0.7433 val_degree_1_accuracy: 0.9082 val_degree_2_accuracy: 0.6919 val_quality_accuracy: 0.7372 val_inversion_accuracy: 0.6658 val_root_accuracy: 0.7895 ``` The output of `analyse_results` ```python {'key': 74.32721613384346, 'degree 1': 90.81771610142015, 'degree 2': 69.19136242785811, 'quality': 73.71765773944621, 'inversion': 66.58452759224434, 'root': 78.9507814019843, 'degree': 62.953115880941574, 'secondary': 3.68271954674221, 'derived root': 65.55346605278517, 'roman': 46.929511704818104, 'roman + inv': 31.787821801439595, 'root coherence': 72.58284157966409, 'd7 no inv': 8.877284595300262} ```

Here is experiment "BPSSynth on BPSSynth-val":

Epoch 28/100
 1/93 [..............................] - ETA: 1:04 - loss: 0.7217 - key_loss: 0.1793 - degree_1_loss: 0.1467 - degree_2_loss: 0.1864 - quality_loss: 0.0988 - inversion_loss: 0.0631 - root_loss: 0.0474 - key_accu 2/93 [..............................] - ETA: 54s - loss: 0.7953 - key_loss: 0.2170 - degree_1_loss: 0.1723 - degree_2_loss: 0.1927 - quality_loss: 0.0929 - inversion_loss: 0.0611 - root_loss: 0.0592 - key_accur 3/93 [..............................] - ETA: 52s - loss: 0.8121 - key_loss: 0.2218 - degree_1_loss: 0.1770 - degree_2_loss: 0.1992 - quality_loss: 0.0929 - inversion_loss: 0.0615 - root_loss: 0.0599 - key_accur 4/93 [>.............................] - ETA: 52s - loss: 0.8234 - key_loss: 0.2224 - degree_1_loss: 0.1841 - degree_2_loss: 0.2004 - quality_loss: 0.0936 - inversion_loss: 0.0621 - root_loss: 0.0607 - key_accur 5/93 [>.............................] - ETA: 51s - loss: 0.8496 - key_loss: 0.2299 - degree_1_loss: 0.1909 - degree_2_loss: 0.2069 - quality_loss: 0.0947 - inversion_loss: 0.0636 - root_loss: 0.0636 - key_accur 6/93 [>.............................] - ETA: 50s - loss: 0.8676 - key_loss: 0.2348 - degree_1_loss: 0.1953 - degree_2_loss: 0.2110 - quality_loss: 0.0956 - inversion_loss: 0.0646 - root_loss: 0.0664 - key_accur 7/93 [=>............................] - ETA: 50s - loss: 0.8801 - key_loss: 0.2382 - degree_1_loss: 0.1985 - degree_2_loss: 0.2151 - quality_loss: 0.0960 - inversion_loss: 0.0649 - root_loss: 0.0675 - key_accur 8/93 [=>............................] - ETA: 49s - loss: 0.8884 - key_loss: 0.2401 - degree_1_loss: 0.2005 - degree_2_loss: 0.2182 - quality_loss: 0.0962 - inversion_loss: 0.0655 - root_loss: 0.0679 - key_accur 9/93 [=>............................] - ETA: 49s - loss: 0.8959 - key_loss: 0.2421 - degree_1_loss: 0.2028 - degree_2_loss: 0.2207 - quality_loss: 0.0961 - inversion_loss: 0.0658 - root_loss: 0.0683 - key_accur10/93 [==>...........................] - ETA: 48s - loss: 0.9033 - key_loss: 0.2448 - degree_1_loss: 0.2043 - degree_2_loss: 0.2234 - quality_loss: 0.0963 - inversion_loss: 0.0662 - root_loss: 0.0684 - key_accur11/93 [==>...........................] - ETA: 48s - loss: 0.9135 - key_loss: 0.2484 - degree_1_loss: 0.2060 - degree_2_loss: 0.2266 - quality_loss: 0.0970 - inversion_loss: 0.0668 - root_loss: 0.0687 - key_accur12/93 [==>...........................] - ETA: 47s - loss: 0.9218 - key_loss: 0.2517 - degree_1_loss: 0.2075 - degree_2_loss: 0.2292 - quality_loss: 0.0975 - inversion_loss: 0.0672 - root_loss: 0.0688 - key_accur13/93 [===>..........................] - ETA: 47s - loss: 0.9291 - key_loss: 0.2543 - degree_1_loss: 0.2090 - degree_2_loss: 0.2313 - quality_loss: 0.0977 - inversion_loss: 0.0676 - root_loss: 0.0691 - key_accur14/93 [===>..........................] - ETA: 46s - loss: 0.9351 - key_loss: 0.2566 - degree_1_loss: 0.2103 - degree_2_loss: 0.2329 - quality_loss: 0.0980 - inversion_loss: 0.0680 - root_loss: 0.0692 - key_accur15/93 [===>..........................] - ETA: 46s - loss: 0.9399 - key_loss: 0.2584 - degree_1_loss: 0.2113 - degree_2_loss: 0.2344 - quality_loss: 0.0983 - inversion_loss: 0.0683 - root_loss: 0.0692 - key_accur16/93 [====>.........................] - ETA: 45s - loss: 0.9441 - key_loss: 0.2600 - degree_1_loss: 0.2120 - degree_2_loss: 0.2357 - quality_loss: 0.0986 - inversion_loss: 0.0687 - root_loss: 0.0691 - key_accur17/93 [====>.........................] - ETA: 44s - loss: 0.9480 - key_loss: 0.2613 - degree_1_loss: 0.2128 - degree_2_loss: 0.2370 - quality_loss: 0.0989 - inversion_loss: 0.0690 - root_loss: 0.0690 - key_accur18/93 [====>.........................] - ETA: 44s - loss: 0.9521 - key_loss: 0.2624 - degree_1_loss: 0.2135 - degree_2_loss: 0.2383 - quality_loss: 0.0993 - inversion_loss: 0.0696 - root_loss: 0.0690 - key_accur19/93 [=====>........................] - ETA: 43s - loss: 0.9568 - key_loss: 0.2637 - degree_1_loss: 0.2146 - degree_2_loss: 0.2397 - quality_loss: 0.0997 - inversion_loss: 0.0701 - root_loss: 0.0690 - key_accur20/93 [=====>........................] - ETA: 43s - loss: 0.9611 - key_loss: 0.2650 - degree_1_loss: 0.2157 - degree_2_loss: 0.2410 - quality_loss: 0.1000 - inversion_loss: 0.0705 - root_loss: 0.0689 - key_accur21/93 [=====>........................] - ETA: 42s - loss: 0.9646 - key_loss: 0.2662 - degree_1_loss: 0.2164 - degree_2_loss: 0.2420 - quality_loss: 0.1003 - inversion_loss: 0.0709 - root_loss: 0.0688 - key_accur22/93 [======>.......................] - ETA: 41s - loss: 0.9677 - key_loss: 0.2673 - degree_1_loss: 0.2170 - degree_2_loss: 0.2429 - quality_loss: 0.1005 - inversion_loss: 0.0713 - root_loss: 0.0687 - key_accur23/93 [======>.......................] - ETA: 41s - loss: 0.9706 - key_loss: 0.2684 - degree_1_loss: 0.2175 - degree_2_loss: 0.2438 - quality_loss: 0.1008 - inversion_loss: 0.0716 - root_loss: 0.0685 - key_accur24/93 [======>.......................] - ETA: 40s - loss: 0.9734 - key_loss: 0.2694 - degree_1_loss: 0.2181 - degree_2_loss: 0.2446 - quality_loss: 0.1010 - inversion_loss: 0.0719 - root_loss: 0.0683 - key_accur25/93 [=======>......................] - ETA: 40s - loss: 0.9760 - key_loss: 0.2704 - degree_1_loss: 0.2187 - degree_2_loss: 0.2454 - quality_loss: 0.1012 - inversion_loss: 0.0722 - root_loss: 0.0682 - key_accur26/93 [=======>......................] - ETA: 39s - loss: 0.9782 - key_loss: 0.2712 - degree_1_loss: 0.2192 - degree_2_loss: 0.2460 - quality_loss: 0.1013 - inversion_loss: 0.0724 - root_loss: 0.0681 - key_accur27/93 [=======>......................] - ETA: 38s - loss: 0.9800 - key_loss: 0.2719 - degree_1_loss: 0.2197 - degree_2_loss: 0.2465 - quality_loss: 0.1015 - inversion_loss: 0.0726 - root_loss: 0.0679 - key_accur28/93 [========>.....................] - ETA: 38s - loss: 0.9819 - key_loss: 0.2726 - degree_1_loss: 0.2201 - degree_2_loss: 0.2471 - quality_loss: 0.1016 - inversion_loss: 0.0727 - root_loss: 0.0678 - key_accur29/93 [========>.....................] - ETA: 37s - loss: 0.9838 - key_loss: 0.2733 - degree_1_loss: 0.2206 - degree_2_loss: 0.2477 - quality_loss: 0.1017 - inversion_loss: 0.0728 - root_loss: 0.0677 - key_accur30/93 [========>.....................] - ETA: 36s - loss: 0.9856 - key_loss: 0.2741 - degree_1_loss: 0.2209 - degree_2_loss: 0.2482 - quality_loss: 0.1019 - inversion_loss: 0.0729 - root_loss: 0.0675 - key_accur31/93 [=========>....................] - ETA: 36s - loss: 0.9871 - key_loss: 0.2748 - degree_1_loss: 0.2213 - degree_2_loss: 0.2486 - quality_loss: 0.1020 - inversion_loss: 0.0730 - root_loss: 0.0674 - key_accur32/93 [=========>....................] - ETA: 35s - loss: 0.9885 - key_loss: 0.2754 - degree_1_loss: 0.2217 - degree_2_loss: 0.2490 - quality_loss: 0.1021 - inversion_loss: 0.0731 - root_loss: 0.0672 - key_accur33/93 [=========>....................] - ETA: 35s - loss: 0.9900 - key_loss: 0.2761 - degree_1_loss: 0.2221 - degree_2_loss: 0.2495 - quality_loss: 0.1022 - inversion_loss: 0.0731 - root_loss: 0.0671 - key_accur34/93 [=========>....................] - ETA: 34s - loss: 0.9914 - key_loss: 0.2767 - degree_1_loss: 0.2225 - degree_2_loss: 0.2499 - quality_loss: 0.1022 - inversion_loss: 0.0731 - root_loss: 0.0669 - key_accur35/93 [==========>...................] - ETA: 33s - loss: 0.9929 - key_loss: 0.2774 - degree_1_loss: 0.2229 - degree_2_loss: 0.2504 - quality_loss: 0.1023 - inversion_loss: 0.0732 - root_loss: 0.0668 - key_accur36/93 [==========>...................] - ETA: 33s - loss: 0.9945 - key_loss: 0.2781 - degree_1_loss: 0.2233 - degree_2_loss: 0.2508 - quality_loss: 0.1024 - inversion_loss: 0.0732 - root_loss: 0.0666 - key_accur37/93 [==========>...................] - ETA: 32s - loss: 0.9962 - key_loss: 0.2789 - degree_1_loss: 0.2238 - degree_2_loss: 0.2512 - quality_loss: 0.1025 - inversion_loss: 0.0733 - root_loss: 0.0665 - key_accur38/93 [===========>..................] - ETA: 32s - loss: 0.9978 - key_loss: 0.2796 - degree_1_loss: 0.2242 - degree_2_loss: 0.2516 - quality_loss: 0.1026 - inversion_loss: 0.0734 - root_loss: 0.0664 - key_accur39/93 [===========>..................] - ETA: 31s - loss: 0.9992 - key_loss: 0.2804 - degree_1_loss: 0.2246 - degree_2_loss: 0.2520 - quality_loss: 0.1026 - inversion_loss: 0.0734 - root_loss: 0.0663 - key_accur40/93 [===========>..................] - ETA: 31s - loss: 1.0007 - key_loss: 0.2811 - degree_1_loss: 0.2249 - degree_2_loss: 0.2523 - quality_loss: 0.1027 - inversion_loss: 0.0734 - root_loss: 0.0662 - key_accur41/93 [============>.................] - ETA: 30s - loss: 1.0020 - key_loss: 0.2819 - degree_1_loss: 0.2253 - degree_2_loss: 0.2527 - quality_loss: 0.1027 - inversion_loss: 0.0734 - root_loss: 0.0660 - key_accur42/93 [============>.................] - ETA: 29s - loss: 1.0032 - key_loss: 0.2825 - degree_1_loss: 0.2257 - degree_2_loss: 0.2530 - quality_loss: 0.1027 - inversion_loss: 0.0734 - root_loss: 0.0659 - key_accur43/93 [============>.................] - ETA: 29s - loss: 1.0043 - key_loss: 0.2831 - degree_1_loss: 0.2260 - degree_2_loss: 0.2533 - quality_loss: 0.1027 - inversion_loss: 0.0734 - root_loss: 0.0658 - key_accur44/93 [=============>................] - ETA: 28s - loss: 1.0053 - key_loss: 0.2836 - degree_1_loss: 0.2263 - degree_2_loss: 0.2535 - quality_loss: 0.1028 - inversion_loss: 0.0734 - root_loss: 0.0657 - key_accur45/93 [=============>................] - ETA: 28s - loss: 1.0062 - key_loss: 0.2841 - degree_1_loss: 0.2266 - degree_2_loss: 0.2537 - quality_loss: 0.1028 - inversion_loss: 0.0734 - root_loss: 0.0656 - key_accur46/93 [=============>................] - ETA: 27s - loss: 1.0071 - key_loss: 0.2846 - degree_1_loss: 0.2269 - degree_2_loss: 0.2540 - quality_loss: 0.1028 - inversion_loss: 0.0734 - root_loss: 0.0655 - key_accur47/93 [==============>...............] - ETA: 26s - loss: 1.0079 - key_loss: 0.2850 - degree_1_loss: 0.2272 - degree_2_loss: 0.2542 - quality_loss: 0.1028 - inversion_loss: 0.0734 - root_loss: 0.0654 - key_accur48/93 [==============>...............] - ETA: 26s - loss: 1.0089 - key_loss: 0.2855 - degree_1_loss: 0.2275 - degree_2_loss: 0.2544 - quality_loss: 0.1029 - inversion_loss: 0.0733 - root_loss: 0.0654 - key_accur49/93 [==============>...............] - ETA: 25s - loss: 1.0099 - key_loss: 0.2859 - degree_1_loss: 0.2279 - degree_2_loss: 0.2546 - quality_loss: 0.1029 - inversion_loss: 0.0733 - root_loss: 0.0653 - key_accur50/93 [===============>..............] - ETA: 25s - loss: 1.0109 - key_loss: 0.2864 - degree_1_loss: 0.2282 - degree_2_loss: 0.2548 - quality_loss: 0.1029 - inversion_loss: 0.0733 - root_loss: 0.0653 - key_accur51/93 [===============>..............] - ETA: 24s - loss: 1.0118 - key_loss: 0.2869 - degree_1_loss: 0.2286 - degree_2_loss: 0.2549 - quality_loss: 0.1030 - inversion_loss: 0.0733 - root_loss: 0.0652 - key_accur52/93 [===============>..............] - ETA: 23s - loss: 1.0127 - key_loss: 0.2873 - degree_1_loss: 0.2289 - degree_2_loss: 0.2551 - quality_loss: 0.1030 - inversion_loss: 0.0732 - root_loss: 0.0652 - key_accur53/93 [================>.............] - ETA: 23s - loss: 1.0134 - key_loss: 0.2877 - degree_1_loss: 0.2292 - degree_2_loss: 0.2551 - quality_loss: 0.1031 - inversion_loss: 0.0732 - root_loss: 0.0651 - key_accur54/93 [================>.............] - ETA: 22s - loss: 1.0142 - key_loss: 0.2881 - degree_1_loss: 0.2295 - degree_2_loss: 0.2552 - quality_loss: 0.1031 - inversion_loss: 0.0732 - root_loss: 0.0651 - key_accur55/93 [================>.............] - ETA: 22s - loss: 1.0148 - key_loss: 0.2885 - degree_1_loss: 0.2297 - degree_2_loss: 0.2553 - quality_loss: 0.1032 - inversion_loss: 0.0731 - root_loss: 0.0650 - key_accur56/93 [=================>............] - ETA: 21s - loss: 1.0154 - key_loss: 0.2888 - degree_1_loss: 0.2300 - degree_2_loss: 0.2554 - quality_loss: 0.1032 - inversion_loss: 0.0731 - root_loss: 0.0650 - key_accur57/93 [=================>............] - ETA: 20s - loss: 1.0161 - key_loss: 0.2892 - degree_1_loss: 0.2302 - degree_2_loss: 0.2555 - quality_loss: 0.1032 - inversion_loss: 0.0730 - root_loss: 0.0649 - key_accur58/93 [=================>............] - ETA: 20s - loss: 1.0167 - key_loss: 0.2895 - degree_1_loss: 0.2305 - degree_2_loss: 0.2555 - quality_loss: 0.1033 - inversion_loss: 0.0730 - root_loss: 0.0649 - key_accur59/93 [==================>...........] - ETA: 19s - loss: 1.0174 - key_loss: 0.2899 - degree_1_loss: 0.2307 - degree_2_loss: 0.2556 - quality_loss: 0.1033 - inversion_loss: 0.0730 - root_loss: 0.0649 - key_accur60/93 [==================>...........] - ETA: 19s - loss: 1.0180 - key_loss: 0.2902 - degree_1_loss: 0.2309 - degree_2_loss: 0.2557 - quality_loss: 0.1034 - inversion_loss: 0.0730 - root_loss: 0.0648 - key_accur61/93 [==================>...........] - ETA: 18s - loss: 1.0187 - key_loss: 0.2906 - degree_1_loss: 0.2312 - degree_2_loss: 0.2557 - quality_loss: 0.1034 - inversion_loss: 0.0730 - root_loss: 0.0648 - key_accur62/93 [===================>..........] - ETA: 18s - loss: 1.0195 - key_loss: 0.2910 - degree_1_loss: 0.2314 - degree_2_loss: 0.2558 - quality_loss: 0.1035 - inversion_loss: 0.0730 - root_loss: 0.0648 - key_accur63/93 [===================>..........] - ETA: 17s - loss: 1.0202 - key_loss: 0.2914 - degree_1_loss: 0.2316 - degree_2_loss: 0.2559 - quality_loss: 0.1036 - inversion_loss: 0.0730 - root_loss: 0.0648 - key_accur64/93 [===================>..........] - ETA: 16s - loss: 1.0209 - key_loss: 0.2917 - degree_1_loss: 0.2318 - degree_2_loss: 0.2560 - quality_loss: 0.1036 - inversion_loss: 0.0730 - root_loss: 0.0647 - key_accur65/93 [===================>..........] - ETA: 16s - loss: 1.0215 - key_loss: 0.2921 - degree_1_loss: 0.2320 - degree_2_loss: 0.2561 - quality_loss: 0.1037 - inversion_loss: 0.0730 - root_loss: 0.0647 - key_accur66/93 [====================>.........] - ETA: 15s - loss: 1.0221 - key_loss: 0.2924 - degree_1_loss: 0.2322 - degree_2_loss: 0.2561 - quality_loss: 0.1038 - inversion_loss: 0.0729 - root_loss: 0.0647 - key_accur67/93 [====================>.........] - ETA: 15s - loss: 1.0226 - key_loss: 0.2927 - degree_1_loss: 0.2323 - degree_2_loss: 0.2562 - quality_loss: 0.1038 - inversion_loss: 0.0729 - root_loss: 0.0647 - key_accur68/93 [====================>.........] - ETA: 14s - loss: 1.0231 - key_loss: 0.2929 - degree_1_loss: 0.2325 - degree_2_loss: 0.2562 - quality_loss: 0.1039 - inversion_loss: 0.0729 - root_loss: 0.0646 - key_accur69/93 [=====================>........] - ETA: 13s - loss: 1.0235 - key_loss: 0.2932 - degree_1_loss: 0.2326 - degree_2_loss: 0.2563 - quality_loss: 0.1040 - inversion_loss: 0.0729 - root_loss: 0.0646 - key_accur70/93 [=====================>........] - ETA: 13s - loss: 1.0240 - key_loss: 0.2935 - degree_1_loss: 0.2328 - degree_2_loss: 0.2563 - quality_loss: 0.1040 - inversion_loss: 0.0728 - root_loss: 0.0645 - key_accur71/93 [=====================>........] - ETA: 12s - loss: 1.0243 - key_loss: 0.2937 - degree_1_loss: 0.2329 - degree_2_loss: 0.2564 - quality_loss: 0.1041 - inversion_loss: 0.0728 - root_loss: 0.0645 - key_accur72/93 [======================>.......] - ETA: 12s - loss: 1.0246 - key_loss: 0.2939 - degree_1_loss: 0.2330 - degree_2_loss: 0.2564 - quality_loss: 0.1041 - inversion_loss: 0.0728 - root_loss: 0.0645 - key_accur73/93 [======================>.......] - ETA: 11s - loss: 1.0249 - key_loss: 0.2941 - degree_1_loss: 0.2331 - degree_2_loss: 0.2564 - quality_loss: 0.1041 - inversion_loss: 0.0727 - root_loss: 0.0644 - key_accur74/93 [======================>.......] - ETA: 11s - loss: 1.0252 - key_loss: 0.2944 - degree_1_loss: 0.2332 - degree_2_loss: 0.2564 - quality_loss: 0.1042 - inversion_loss: 0.0727 - root_loss: 0.0644 - key_accur75/93 [=======================>......] - ETA: 10s - loss: 1.0255 - key_loss: 0.2945 - degree_1_loss: 0.2333 - degree_2_loss: 0.2564 - quality_loss: 0.1042 - inversion_loss: 0.0727 - root_loss: 0.0643 - key_accur76/93 [=======================>......] - ETA: 9s - loss: 1.0257 - key_loss: 0.2947 - degree_1_loss: 0.2334 - degree_2_loss: 0.2564 - quality_loss: 0.1042 - inversion_loss: 0.0726 - root_loss: 0.0643 - key_accura77/93 [=======================>......] - ETA: 9s - loss: 1.0258 - key_loss: 0.2948 - degree_1_loss: 0.2335 - degree_2_loss: 0.2564 - quality_loss: 0.1042 - inversion_loss: 0.0726 - root_loss: 0.0642 - key_accura78/93 [========================>.....] - ETA: 8s - loss: 1.0260 - key_loss: 0.2950 - degree_1_loss: 0.2336 - degree_2_loss: 0.2564 - quality_loss: 0.1042 - inversion_loss: 0.0726 - root_loss: 0.0642 - key_accura79/93 [========================>.....] - ETA: 8s - loss: 1.0262 - key_loss: 0.2951 - degree_1_loss: 0.2337 - degree_2_loss: 0.2563 - quality_loss: 0.1043 - inversion_loss: 0.0725 - root_loss: 0.0641 - key_accura80/93 [========================>.....] - ETA: 7s - loss: 1.0263 - key_loss: 0.2953 - degree_1_loss: 0.2338 - degree_2_loss: 0.2563 - quality_loss: 0.1043 - inversion_loss: 0.0725 - root_loss: 0.0641 - key_accura81/93 [=========================>....] - ETA: 6s - loss: 1.0265 - key_loss: 0.2954 - degree_1_loss: 0.2339 - degree_2_loss: 0.2563 - quality_loss: 0.1043 - inversion_loss: 0.0725 - root_loss: 0.0641 - key_accura82/93 [=========================>....] - ETA: 6s - loss: 1.0267 - key_loss: 0.2956 - degree_1_loss: 0.2340 - degree_2_loss: 0.2563 - quality_loss: 0.1043 - inversion_loss: 0.0725 - root_loss: 0.0640 - key_accura83/93 [=========================>....] - ETA: 5s - loss: 1.0269 - key_loss: 0.2958 - degree_1_loss: 0.2341 - degree_2_loss: 0.2563 - quality_loss: 0.1044 - inversion_loss: 0.0724 - root_loss: 0.0640 - key_accura84/93 [==========================>...] - ETA: 5s - loss: 1.0271 - key_loss: 0.2959 - degree_1_loss: 0.2342 - degree_2_loss: 0.2563 - quality_loss: 0.1044 - inversion_loss: 0.0724 - root_loss: 0.0639 - key_accura85/93 [==========================>...] - ETA: 4s - loss: 1.0272 - key_loss: 0.2961 - degree_1_loss: 0.2343 - degree_2_loss: 0.2562 - quality_loss: 0.1044 - inversion_loss: 0.0724 - root_loss: 0.0639 - key_accura86/93 [==========================>...] - ETA: 4s - loss: 1.0274 - key_loss: 0.2962 - degree_1_loss: 0.2344 - degree_2_loss: 0.2562 - quality_loss: 0.1044 - inversion_loss: 0.0724 - root_loss: 0.0638 - key_accura87/93 [===========================>..] - ETA: 3s - loss: 1.0275 - key_loss: 0.2963 - degree_1_loss: 0.2344 - degree_2_loss: 0.2562 - quality_loss: 0.1044 - inversion_loss: 0.0723 - root_loss: 0.0638 - key_accura88/93 [===========================>..] - ETA: 2s - loss: 1.0276 - key_loss: 0.2965 - degree_1_loss: 0.2345 - degree_2_loss: 0.2561 - quality_loss: 0.1044 - inversion_loss: 0.0723 - root_loss: 0.0638 - key_accura89/93 [===========================>..] - ETA: 2s - loss: 1.0276 - key_loss: 0.2966 - degree_1_loss: 0.2346 - degree_2_loss: 0.2561 - quality_loss: 0.1044 - inversion_loss: 0.0723 - root_loss: 0.0637 - key_accura90/93 [============================>.] - ETA: 1s - loss: 1.0277 - key_loss: 0.2967 - degree_1_loss: 0.2346 - degree_2_loss: 0.2560 - quality_loss: 0.1044 - inversion_loss: 0.0722 - root_loss: 0.0637 - key_accura91/93 [============================>.] - ETA: 1s - loss: 1.0278 - key_loss: 0.2968 - degree_1_loss: 0.2347 - degree_2_loss: 0.2560 - quality_loss: 0.1044 - inversion_loss: 0.0722 - root_loss: 0.0636 - key_accura92/93 [============================>.] - ETA: 0s - loss: 1.0279 - key_loss: 0.2970 - degree_1_loss: 0.2348 - degree_2_loss: 0.2560 - quality_loss: 0.1044 - inversion_loss: 0.0722 - root_loss: 0.0636 - key_accura93/93 [==============================] - ETA: 0s - loss: 1.0280 - key_loss: 0.2971 - degree_1_loss: 0.2348 - degree_2_loss: 0.2559 - quality_loss: 0.1044 - inversion_loss: 0.0721 - root_loss: 0.0636 - key_accura93/93 [==============================] - 54s 585ms/step - loss: 1.0282 - key_loss: 0.2973 - degree_1_loss: 0.2349 - degree_2_loss: 0.2559 - quality_loss: 0.1045 - inversion_loss: 0.0721 - root_loss: 0.0635 - key_accuracy: 0.9084 - degree_1_accuracy: 0.9271 - degree_2_accuracy: 0.9171 - quality_accuracy: 0.9675 - inversion_accuracy: 0.9768 - root_accuracy: 0.9862 - val_loss: 1.5637 - val_key_loss: 0.6049 - val_degree_1_loss: 0.2979 - val_degree_2_loss: 0.4184 - val_quality_loss: 0.1339 - val_inversion_loss: 0.0678 - val_root_loss: 0.0408 - val_key_accuracy: 0.7984 - val_degree_1_accuracy: 0.9168 - val_degree_2_accuracy: 0.8619 - val_quality_accuracy: 0.9619 - val_inversion_accuracy: 0.9823 - val_root_accuracy: 0.9908

loss: 1.0282
key_loss: 0.2973
degree_1_loss: 0.2349
degree_2_loss: 0.2559
quality_loss: 0.1045
inversion_loss: 0.0721
root_loss: 0.0635

key_accuracy: 0.9084
degree_1_accuracy: 0.9271
degree_2_accuracy: 0.9171
quality_accuracy: 0.9675
inversion_accuracy: 0.9768
root_accuracy: 0.9862

val_loss: 1.5637
val_key_loss: 0.6049
val_degree_1_loss: 0.2979
val_degree_2_loss: 0.4184
val_quality_loss: 0.1339
val_inversion_loss: 0.0678
val_root_loss: 0.0408

val_key_accuracy: 0.7984
val_degree_1_accuracy: 0.9168
val_degree_2_accuracy: 0.8619
val_quality_accuracy: 0.9619
val_inversion_accuracy: 0.9823
val_root_accuracy: 0.9908

The output of analyse_results:

{
    "key": 79.83806258662193,
    "degree 1": 91.67700051061347,
    "degree 2": 86.19155299438326,
    "quality": 96.19228244219126,
    "inversion": 98.2347363046174,
    "root": 99.08089576190824,
    "degree": 78.40105040484353,
    "secondary": 23.37565347274085,
    "derived root": 85.76847326573784,
    "roman": 69.70603253337224,
    "roman + inv": 69.37778101976804,
    "root coherence": 85.6882340068568,
    "d7 no inv": 42.556390977443606,
}

Here is experiment "BPSSynth on BPS-val"

Epoch 7/100
 1/93 [..............................] - ETA: 58s - loss: 3.2676 - key_loss: 1.0581 - degree_1_loss: 0.4122 - degree_2_loss: 0.9053 - quality_loss: 0.3102 - inversion_loss: 0.3649 - root_loss: 0.2170 - key_accur 2/93 [..............................] - ETA: 53s - loss: 3.0705 - key_loss: 0.9866 - degree_1_loss: 0.4281 - degree_2_loss: 0.7974 - quality_loss: 0.2938 - inversion_loss: 0.3436 - root_loss: 0.2211 - key_accur 3/93 [..............................] - ETA: 53s - loss: 2.9786 - key_loss: 0.9564 - degree_1_loss: 0.4281 - degree_2_loss: 0.7528 - quality_loss: 0.2822 - inversion_loss: 0.3365 - root_loss: 0.2225 - key_accur 4/93 [>.............................] - ETA: 52s - loss: 2.9132 - key_loss: 0.9290 - degree_1_loss: 0.4255 - degree_2_loss: 0.7231 - quality_loss: 0.2772 - inversion_loss: 0.3313 - root_loss: 0.2270 - key_accur 5/93 [>.............................] - ETA: 51s - loss: 2.8761 - key_loss: 0.9139 - degree_1_loss: 0.4247 - degree_2_loss: 0.7035 - quality_loss: 0.2750 - inversion_loss: 0.3277 - root_loss: 0.2314 - key_accur 6/93 [>.............................] - ETA: 50s - loss: 2.8476 - key_loss: 0.9042 - degree_1_loss: 0.4236 - degree_2_loss: 0.6878 - quality_loss: 0.2735 - inversion_loss: 0.3243 - root_loss: 0.2342 - key_accur 7/93 [=>............................] - ETA: 50s - loss: 2.8277 - key_loss: 0.9008 - degree_1_loss: 0.4219 - degree_2_loss: 0.6745 - quality_loss: 0.2722 - inversion_loss: 0.3222 - root_loss: 0.2362 - key_accur 8/93 [=>............................] - ETA: 49s - loss: 2.8089 - key_loss: 0.8977 - degree_1_loss: 0.4190 - degree_2_loss: 0.6654 - quality_loss: 0.2707 - inversion_loss: 0.3201 - root_loss: 0.2360 - key_accur 9/93 [=>............................] - ETA: 48s - loss: 2.7934 - key_loss: 0.8948 - degree_1_loss: 0.4159 - degree_2_loss: 0.6583 - quality_loss: 0.2698 - inversion_loss: 0.3187 - root_loss: 0.2359 - key_accur10/93 [==>...........................] - ETA: 48s - loss: 2.7858 - key_loss: 0.8931 - degree_1_loss: 0.4136 - degree_2_loss: 0.6538 - quality_loss: 0.2705 - inversion_loss: 0.3185 - root_loss: 0.2361 - key_accur11/93 [==>...........................] - ETA: 47s - loss: 2.7777 - key_loss: 0.8906 - degree_1_loss: 0.4107 - degree_2_loss: 0.6500 - quality_loss: 0.2716 - inversion_loss: 0.3184 - root_loss: 0.2363 - key_accur12/93 [==>...........................] - ETA: 47s - loss: 2.7688 - key_loss: 0.8877 - degree_1_loss: 0.4080 - degree_2_loss: 0.6462 - quality_loss: 0.2723 - inversion_loss: 0.3184 - root_loss: 0.2362 - key_accur13/93 [===>..........................] - ETA: 46s - loss: 2.7634 - key_loss: 0.8863 - degree_1_loss: 0.4054 - degree_2_loss: 0.6440 - quality_loss: 0.2735 - inversion_loss: 0.3183 - root_loss: 0.2359 - key_accur14/93 [===>..........................] - ETA: 45s - loss: 2.7610 - key_loss: 0.8863 - degree_1_loss: 0.4033 - degree_2_loss: 0.6422 - quality_loss: 0.2747 - inversion_loss: 0.3184 - root_loss: 0.2361 - key_accur15/93 [===>..........................] - ETA: 45s - loss: 2.7584 - key_loss: 0.8864 - degree_1_loss: 0.4015 - degree_2_loss: 0.6408 - quality_loss: 0.2754 - inversion_loss: 0.3182 - root_loss: 0.2361 - key_accur16/93 [====>.........................] - ETA: 44s - loss: 2.7560 - key_loss: 0.8862 - degree_1_loss: 0.3997 - degree_2_loss: 0.6396 - quality_loss: 0.2761 - inversion_loss: 0.3181 - root_loss: 0.2363 - key_accur17/93 [====>.........................] - ETA: 43s - loss: 2.7536 - key_loss: 0.8858 - degree_1_loss: 0.3982 - degree_2_loss: 0.6382 - quality_loss: 0.2767 - inversion_loss: 0.3183 - root_loss: 0.2363 - key_accur18/93 [====>.........................] - ETA: 43s - loss: 2.7496 - key_loss: 0.8846 - degree_1_loss: 0.3966 - degree_2_loss: 0.6366 - quality_loss: 0.2773 - inversion_loss: 0.3184 - root_loss: 0.2361 - key_accur19/93 [=====>........................] - ETA: 42s - loss: 2.7456 - key_loss: 0.8835 - degree_1_loss: 0.3950 - degree_2_loss: 0.6351 - quality_loss: 0.2778 - inversion_loss: 0.3182 - root_loss: 0.2359 - key_accur20/93 [=====>........................] - ETA: 42s - loss: 2.7414 - key_loss: 0.8826 - degree_1_loss: 0.3937 - degree_2_loss: 0.6333 - quality_loss: 0.2781 - inversion_loss: 0.3180 - root_loss: 0.2356 - key_accur21/93 [=====>........................] - ETA: 41s - loss: 2.7382 - key_loss: 0.8821 - degree_1_loss: 0.3925 - degree_2_loss: 0.6319 - quality_loss: 0.2785 - inversion_loss: 0.3178 - root_loss: 0.2354 - key_accur22/93 [======>.......................] - ETA: 40s - loss: 2.7351 - key_loss: 0.8814 - degree_1_loss: 0.3915 - degree_2_loss: 0.6303 - quality_loss: 0.2789 - inversion_loss: 0.3176 - root_loss: 0.2354 - key_accur23/93 [======>.......................] - ETA: 40s - loss: 2.7322 - key_loss: 0.8809 - degree_1_loss: 0.3906 - degree_2_loss: 0.6289 - quality_loss: 0.2792 - inversion_loss: 0.3173 - root_loss: 0.2354 - key_accur24/93 [======>.......................] - ETA: 39s - loss: 2.7303 - key_loss: 0.8804 - degree_1_loss: 0.3898 - degree_2_loss: 0.6277 - quality_loss: 0.2797 - inversion_loss: 0.3172 - root_loss: 0.2356 - key_accur25/93 [=======>......................] - ETA: 39s - loss: 2.7273 - key_loss: 0.8795 - degree_1_loss: 0.3889 - degree_2_loss: 0.6264 - quality_loss: 0.2801 - inversion_loss: 0.3169 - root_loss: 0.2356 - key_accur26/93 [=======>......................] - ETA: 38s - loss: 2.7243 - key_loss: 0.8785 - degree_1_loss: 0.3880 - degree_2_loss: 0.6250 - quality_loss: 0.2804 - inversion_loss: 0.3168 - root_loss: 0.2356 - key_accur27/93 [=======>......................] - ETA: 38s - loss: 2.7216 - key_loss: 0.8778 - degree_1_loss: 0.3873 - degree_2_loss: 0.6239 - quality_loss: 0.2806 - inversion_loss: 0.3166 - root_loss: 0.2355 - key_accur28/93 [========>.....................] - ETA: 37s - loss: 2.7194 - key_loss: 0.8772 - degree_1_loss: 0.3866 - degree_2_loss: 0.6230 - quality_loss: 0.2808 - inversion_loss: 0.3164 - root_loss: 0.2354 - key_accur29/93 [========>.....................] - ETA: 36s - loss: 2.7171 - key_loss: 0.8765 - degree_1_loss: 0.3859 - degree_2_loss: 0.6221 - quality_loss: 0.2810 - inversion_loss: 0.3164 - root_loss: 0.2353 - key_accur30/93 [========>.....................] - ETA: 36s - loss: 2.7152 - key_loss: 0.8759 - degree_1_loss: 0.3853 - degree_2_loss: 0.6212 - quality_loss: 0.2812 - inversion_loss: 0.3163 - root_loss: 0.2353 - key_accur31/93 [=========>....................] - ETA: 35s - loss: 2.7130 - key_loss: 0.8752 - degree_1_loss: 0.3847 - degree_2_loss: 0.6203 - quality_loss: 0.2813 - inversion_loss: 0.3162 - root_loss: 0.2353 - key_accur32/93 [=========>....................] - ETA: 35s - loss: 2.7110 - key_loss: 0.8745 - degree_1_loss: 0.3842 - degree_2_loss: 0.6195 - quality_loss: 0.2814 - inversion_loss: 0.3161 - root_loss: 0.2352 - key_accur33/93 [=========>....................] - ETA: 34s - loss: 2.7088 - key_loss: 0.8739 - degree_1_loss: 0.3838 - degree_2_loss: 0.6186 - quality_loss: 0.2814 - inversion_loss: 0.3161 - root_loss: 0.2351 - key_accur34/93 [=========>....................] - ETA: 34s - loss: 2.7065 - key_loss: 0.8731 - degree_1_loss: 0.3833 - degree_2_loss: 0.6177 - quality_loss: 0.2814 - inversion_loss: 0.3160 - root_loss: 0.2351 - key_accur35/93 [==========>...................] - ETA: 33s - loss: 2.7040 - key_loss: 0.8723 - degree_1_loss: 0.3828 - degree_2_loss: 0.6168 - quality_loss: 0.2813 - inversion_loss: 0.3159 - root_loss: 0.2349 - key_accur36/93 [==========>...................] - ETA: 32s - loss: 2.7014 - key_loss: 0.8714 - degree_1_loss: 0.3823 - degree_2_loss: 0.6159 - quality_loss: 0.2812 - inversion_loss: 0.3159 - root_loss: 0.2348 - key_accur37/93 [==========>...................] - ETA: 32s - loss: 2.6989 - key_loss: 0.8706 - degree_1_loss: 0.3818 - degree_2_loss: 0.6151 - quality_loss: 0.2810 - inversion_loss: 0.3158 - root_loss: 0.2347 - key_accur38/93 [===========>..................] - ETA: 31s - loss: 2.6966 - key_loss: 0.8697 - degree_1_loss: 0.3813 - degree_2_loss: 0.6143 - quality_loss: 0.2810 - inversion_loss: 0.3158 - root_loss: 0.2345 - key_accur39/93 [===========>..................] - ETA: 31s - loss: 2.6944 - key_loss: 0.8690 - degree_1_loss: 0.3809 - degree_2_loss: 0.6134 - quality_loss: 0.2809 - inversion_loss: 0.3158 - root_loss: 0.2344 - key_accur40/93 [===========>..................] - ETA: 30s - loss: 2.6926 - key_loss: 0.8685 - degree_1_loss: 0.3805 - degree_2_loss: 0.6127 - quality_loss: 0.2809 - inversion_loss: 0.3158 - root_loss: 0.2343 - key_accur41/93 [============>.................] - ETA: 29s - loss: 2.6906 - key_loss: 0.8678 - degree_1_loss: 0.3801 - degree_2_loss: 0.6119 - quality_loss: 0.2808 - inversion_loss: 0.3158 - root_loss: 0.2342 - key_accur42/93 [============>.................] - ETA: 29s - loss: 2.6892 - key_loss: 0.8673 - degree_1_loss: 0.3798 - degree_2_loss: 0.6114 - quality_loss: 0.2808 - inversion_loss: 0.3158 - root_loss: 0.2341 - key_accur43/93 [============>.................] - ETA: 28s - loss: 2.6879 - key_loss: 0.8668 - degree_1_loss: 0.3794 - degree_2_loss: 0.6108 - quality_loss: 0.2809 - inversion_loss: 0.3158 - root_loss: 0.2341 - key_accur44/93 [=============>................] - ETA: 28s - loss: 2.6865 - key_loss: 0.8663 - degree_1_loss: 0.3791 - degree_2_loss: 0.6103 - quality_loss: 0.2810 - inversion_loss: 0.3158 - root_loss: 0.2340 - key_accur45/93 [=============>................] - ETA: 27s - loss: 2.6851 - key_loss: 0.8658 - degree_1_loss: 0.3787 - degree_2_loss: 0.6098 - quality_loss: 0.2810 - inversion_loss: 0.3158 - root_loss: 0.2340 - key_accur46/93 [=============>................] - ETA: 27s - loss: 2.6836 - key_loss: 0.8652 - degree_1_loss: 0.3784 - degree_2_loss: 0.6094 - quality_loss: 0.2810 - inversion_loss: 0.3158 - root_loss: 0.2339 - key_accur47/93 [==============>...............] - ETA: 26s - loss: 2.6822 - key_loss: 0.8646 - degree_1_loss: 0.3781 - degree_2_loss: 0.6089 - quality_loss: 0.2811 - inversion_loss: 0.3158 - root_loss: 0.2338 - key_accur48/93 [==============>...............] - ETA: 25s - loss: 2.6810 - key_loss: 0.8641 - degree_1_loss: 0.3778 - degree_2_loss: 0.6085 - quality_loss: 0.2811 - inversion_loss: 0.3158 - root_loss: 0.2338 - key_accur49/93 [==============>...............] - ETA: 25s - loss: 2.6797 - key_loss: 0.8636 - degree_1_loss: 0.3775 - degree_2_loss: 0.6081 - quality_loss: 0.2811 - inversion_loss: 0.3157 - root_loss: 0.2337 - key_accur50/93 [===============>..............] - ETA: 24s - loss: 2.6783 - key_loss: 0.8630 - degree_1_loss: 0.3772 - degree_2_loss: 0.6076 - quality_loss: 0.2811 - inversion_loss: 0.3157 - root_loss: 0.2336 - key_accur51/93 [===============>..............] - ETA: 24s - loss: 2.6771 - key_loss: 0.8625 - degree_1_loss: 0.3768 - degree_2_loss: 0.6073 - quality_loss: 0.2812 - inversion_loss: 0.3157 - root_loss: 0.2336 - key_accur52/93 [===============>..............] - ETA: 23s - loss: 2.6757 - key_loss: 0.8619 - degree_1_loss: 0.3765 - degree_2_loss: 0.6069 - quality_loss: 0.2812 - inversion_loss: 0.3157 - root_loss: 0.2335 - key_accur53/93 [================>.............] - ETA: 22s - loss: 2.6744 - key_loss: 0.8614 - degree_1_loss: 0.3762 - degree_2_loss: 0.6065 - quality_loss: 0.2812 - inversion_loss: 0.3157 - root_loss: 0.2334 - key_accur54/93 [================>.............] - ETA: 22s - loss: 2.6731 - key_loss: 0.8609 - degree_1_loss: 0.3760 - degree_2_loss: 0.6062 - quality_loss: 0.2812 - inversion_loss: 0.3156 - root_loss: 0.2333 - key_accur55/93 [================>.............] - ETA: 21s - loss: 2.6719 - key_loss: 0.8603 - degree_1_loss: 0.3757 - degree_2_loss: 0.6059 - quality_loss: 0.2812 - inversion_loss: 0.3156 - root_loss: 0.2332 - key_accur56/93 [=================>............] - ETA: 21s - loss: 2.6706 - key_loss: 0.8598 - degree_1_loss: 0.3754 - degree_2_loss: 0.6056 - quality_loss: 0.2811 - inversion_loss: 0.3156 - root_loss: 0.2331 - key_accur57/93 [=================>............] - ETA: 20s - loss: 2.6694 - key_loss: 0.8593 - degree_1_loss: 0.3752 - degree_2_loss: 0.6053 - quality_loss: 0.2811 - inversion_loss: 0.3155 - root_loss: 0.2330 - key_accur58/93 [=================>............] - ETA: 20s - loss: 2.6681 - key_loss: 0.8587 - degree_1_loss: 0.3749 - degree_2_loss: 0.6050 - quality_loss: 0.2811 - inversion_loss: 0.3155 - root_loss: 0.2330 - key_accur59/93 [==================>...........] - ETA: 19s - loss: 2.6669 - key_loss: 0.8582 - degree_1_loss: 0.3746 - degree_2_loss: 0.6047 - quality_loss: 0.2811 - inversion_loss: 0.3154 - root_loss: 0.2329 - key_accur60/93 [==================>...........] - ETA: 18s - loss: 2.6656 - key_loss: 0.8577 - degree_1_loss: 0.3744 - degree_2_loss: 0.6044 - quality_loss: 0.2810 - inversion_loss: 0.3154 - root_loss: 0.2328 - key_accur61/93 [==================>...........] - ETA: 18s - loss: 2.6643 - key_loss: 0.8572 - degree_1_loss: 0.3741 - degree_2_loss: 0.6041 - quality_loss: 0.2810 - inversion_loss: 0.3153 - root_loss: 0.2327 - key_accur62/93 [===================>..........] - ETA: 17s - loss: 2.6632 - key_loss: 0.8567 - degree_1_loss: 0.3739 - degree_2_loss: 0.6038 - quality_loss: 0.2809 - inversion_loss: 0.3152 - root_loss: 0.2327 - key_accur63/93 [===================>..........] - ETA: 17s - loss: 2.6621 - key_loss: 0.8563 - degree_1_loss: 0.3736 - degree_2_loss: 0.6035 - quality_loss: 0.2809 - inversion_loss: 0.3152 - root_loss: 0.2326 - key_accur64/93 [===================>..........] - ETA: 16s - loss: 2.6609 - key_loss: 0.8558 - degree_1_loss: 0.3734 - degree_2_loss: 0.6032 - quality_loss: 0.2809 - inversion_loss: 0.3151 - root_loss: 0.2325 - key_accur65/93 [===================>..........] - ETA: 16s - loss: 2.6596 - key_loss: 0.8553 - degree_1_loss: 0.3731 - degree_2_loss: 0.6029 - quality_loss: 0.2809 - inversion_loss: 0.3151 - root_loss: 0.2324 - key_accur66/93 [====================>.........] - ETA: 15s - loss: 2.6582 - key_loss: 0.8547 - degree_1_loss: 0.3728 - degree_2_loss: 0.6026 - quality_loss: 0.2808 - inversion_loss: 0.3150 - root_loss: 0.2323 - key_accur67/93 [====================>.........] - ETA: 14s - loss: 2.6570 - key_loss: 0.8543 - degree_1_loss: 0.3726 - degree_2_loss: 0.6023 - quality_loss: 0.2808 - inversion_loss: 0.3150 - root_loss: 0.2322 - key_accur68/93 [====================>.........] - ETA: 14s - loss: 2.6559 - key_loss: 0.8538 - degree_1_loss: 0.3723 - degree_2_loss: 0.6020 - quality_loss: 0.2808 - inversion_loss: 0.3149 - root_loss: 0.2321 - key_accur69/93 [=====================>........] - ETA: 13s - loss: 2.6548 - key_loss: 0.8533 - degree_1_loss: 0.3721 - degree_2_loss: 0.6017 - quality_loss: 0.2808 - inversion_loss: 0.3148 - root_loss: 0.2320 - key_accur70/93 [=====================>........] - ETA: 13s - loss: 2.6536 - key_loss: 0.8528 - degree_1_loss: 0.3718 - degree_2_loss: 0.6015 - quality_loss: 0.2807 - inversion_loss: 0.3148 - root_loss: 0.2320 - key_accur71/93 [=====================>........] - ETA: 12s - loss: 2.6526 - key_loss: 0.8524 - degree_1_loss: 0.3716 - degree_2_loss: 0.6012 - quality_loss: 0.2807 - inversion_loss: 0.3147 - root_loss: 0.2319 - key_accur72/93 [======================>.......] - ETA: 12s - loss: 2.6516 - key_loss: 0.8520 - degree_1_loss: 0.3714 - degree_2_loss: 0.6010 - quality_loss: 0.2807 - inversion_loss: 0.3147 - root_loss: 0.2318 - key_accur73/93 [======================>.......] - ETA: 11s - loss: 2.6505 - key_loss: 0.8516 - degree_1_loss: 0.3711 - degree_2_loss: 0.6007 - quality_loss: 0.2807 - inversion_loss: 0.3146 - root_loss: 0.2317 - key_accur74/93 [======================>.......] - ETA: 10s - loss: 2.6494 - key_loss: 0.8512 - degree_1_loss: 0.3709 - degree_2_loss: 0.6005 - quality_loss: 0.2806 - inversion_loss: 0.3145 - root_loss: 0.2317 - key_accur75/93 [=======================>......] - ETA: 10s - loss: 2.6484 - key_loss: 0.8508 - degree_1_loss: 0.3707 - degree_2_loss: 0.6003 - quality_loss: 0.2806 - inversion_loss: 0.3145 - root_loss: 0.2316 - key_accur76/93 [=======================>......] - ETA: 9s - loss: 2.6473 - key_loss: 0.8504 - degree_1_loss: 0.3704 - degree_2_loss: 0.6000 - quality_loss: 0.2806 - inversion_loss: 0.3144 - root_loss: 0.2315 - key_accura77/93 [=======================>......] - ETA: 9s - loss: 2.6462 - key_loss: 0.8500 - degree_1_loss: 0.3702 - degree_2_loss: 0.5998 - quality_loss: 0.2806 - inversion_loss: 0.3143 - root_loss: 0.2314 - key_accura78/93 [========================>.....] - ETA: 8s - loss: 2.6452 - key_loss: 0.8496 - degree_1_loss: 0.3699 - degree_2_loss: 0.5996 - quality_loss: 0.2805 - inversion_loss: 0.3142 - root_loss: 0.2313 - key_accura79/93 [========================>.....] - ETA: 8s - loss: 2.6442 - key_loss: 0.8492 - degree_1_loss: 0.3697 - degree_2_loss: 0.5993 - quality_loss: 0.2805 - inversion_loss: 0.3142 - root_loss: 0.2312 - key_accura80/93 [========================>.....] - ETA: 7s - loss: 2.6431 - key_loss: 0.8488 - degree_1_loss: 0.3694 - degree_2_loss: 0.5991 - quality_loss: 0.2805 - inversion_loss: 0.3141 - root_loss: 0.2311 - key_accura81/93 [=========================>....] - ETA: 6s - loss: 2.6420 - key_loss: 0.8484 - degree_1_loss: 0.3692 - degree_2_loss: 0.5989 - quality_loss: 0.2804 - inversion_loss: 0.3140 - root_loss: 0.2310 - key_accura82/93 [=========================>....] - ETA: 6s - loss: 2.6409 - key_loss: 0.8480 - degree_1_loss: 0.3690 - degree_2_loss: 0.5987 - quality_loss: 0.2804 - inversion_loss: 0.3139 - root_loss: 0.2309 - key_accura83/93 [=========================>....] - ETA: 5s - loss: 2.6398 - key_loss: 0.8476 - degree_1_loss: 0.3688 - degree_2_loss: 0.5985 - quality_loss: 0.2804 - inversion_loss: 0.3138 - root_loss: 0.2308 - key_accura84/93 [==========================>...] - ETA: 5s - loss: 2.6387 - key_loss: 0.8472 - degree_1_loss: 0.3685 - degree_2_loss: 0.5982 - quality_loss: 0.2803 - inversion_loss: 0.3137 - root_loss: 0.2307 - key_accura85/93 [==========================>...] - ETA: 4s - loss: 2.6376 - key_loss: 0.8468 - degree_1_loss: 0.3683 - degree_2_loss: 0.5980 - quality_loss: 0.2803 - inversion_loss: 0.3136 - root_loss: 0.2305 - key_accura86/93 [==========================>...] - ETA: 4s - loss: 2.6364 - key_loss: 0.8464 - degree_1_loss: 0.3681 - degree_2_loss: 0.5978 - quality_loss: 0.2802 - inversion_loss: 0.3135 - root_loss: 0.2304 - key_accura87/93 [===========================>..] - ETA: 3s - loss: 2.6353 - key_loss: 0.8460 - degree_1_loss: 0.3678 - degree_2_loss: 0.5976 - quality_loss: 0.2802 - inversion_loss: 0.3134 - root_loss: 0.2303 - key_accura88/93 [===========================>..] - ETA: 2s - loss: 2.6341 - key_loss: 0.8456 - degree_1_loss: 0.3676 - degree_2_loss: 0.5974 - quality_loss: 0.2801 - inversion_loss: 0.3133 - root_loss: 0.2302 - key_accura89/93 [===========================>..] - ETA: 2s - loss: 2.6330 - key_loss: 0.8452 - degree_1_loss: 0.3674 - degree_2_loss: 0.5972 - quality_loss: 0.2801 - inversion_loss: 0.3132 - root_loss: 0.2300 - key_accura90/93 [============================>.] - ETA: 1s - loss: 2.6320 - key_loss: 0.8448 - degree_1_loss: 0.3671 - degree_2_loss: 0.5969 - quality_loss: 0.2800 - inversion_loss: 0.3131 - root_loss: 0.2299 - key_accura91/93 [============================>.] - ETA: 1s - loss: 2.6310 - key_loss: 0.8445 - degree_1_loss: 0.3669 - degree_2_loss: 0.5968 - quality_loss: 0.2800 - inversion_loss: 0.3130 - root_loss: 0.2298 - key_accura92/93 [============================>.] - ETA: 0s - loss: 2.6300 - key_loss: 0.8441 - degree_1_loss: 0.3667 - degree_2_loss: 0.5966 - quality_loss: 0.2800 - inversion_loss: 0.3129 - root_loss: 0.2297 - key_accura93/93 [==============================] - ETA: 0s - loss: 2.6290 - key_loss: 0.8438 - degree_1_loss: 0.3665 - degree_2_loss: 0.5964 - quality_loss: 0.2799 - inversion_loss: 0.3129 - root_loss: 0.2296 - key_accura93/93 [==============================] - 54s 578ms/step - loss: 2.6281 - key_loss: 0.8434 - degree_1_loss: 0.3663 - degree_2_loss: 0.5962 - quality_loss: 0.2799 - inversion_loss: 0.3128 - root_loss: 0.2295 - key_accuracy: 0.7573 - degree_1_accuracy: 0.9009 - degree_2_accuracy: 0.8019 - quality_accuracy: 0.9170 - inversion_accuracy: 0.8852 - root_accuracy: 0.9524 - val_loss: 7.7077 - val_key_loss: 1.3982 - val_degree_1_loss: 0.4216 - val_degree_2_loss: 1.6801 - val_quality_loss: 1.9661 - val_inversion_loss: 1.1834 - val_root_loss: 1.0583 - val_key_accuracy: 0.5508 - val_degree_1_accuracy: 0.9060 - val_degree_2_accuracy: 0.5267 - val_quality_accuracy: 0.5428 - val_inversion_accuracy: 0.6175 - val_root_accuracy: 0.6862

loss: 2.6281
key_loss: 0.8434
degree_1_loss: 0.3663
degree_2_loss: 0.5962
quality_loss: 0.2799
inversion_loss: 0.3128
root_loss: 0.2295

key_accuracy: 0.7573
degree_1_accuracy: 0.9009
degree_2_accuracy: 0.8019
quality_accuracy: 0.9170
inversion_accuracy: 0.8852
root_accuracy: 0.9524

val_loss: 7.7077
val_key_loss: 1.3982
val_degree_1_loss: 0.4216
val_degree_2_loss: 1.6801
val_quality_loss: 1.9661
val_inversion_loss: 1.1834
val_root_loss: 1.0583

val_key_accuracy: 0.5508
val_degree_1_accuracy: 0.9060
val_degree_2_accuracy: 0.5267
val_quality_accuracy: 0.5428
val_inversion_accuracy: 0.6175
val_root_accuracy: 0.6862

The output of analyse_results:

{'key': 55.080734063938785, 
'degree 1': 90.59723753323391, 
'degree 2': 52.6749238052007, 
'quality': 54.28312042020621, 
'inversion': 61.75345308345762, 
'root': 68.62071201608197, 
'degree': 48.08378185591077, 
'secondary': 1.6288951841359773, 
'derived root': 52.73977044290253, 
'roman': 28.059140133584073, 
'roman + inv': 18.75364762337073, 
'root coherence': 59.931262564036054, 
'd7 no inv': 7.9634464751958225}

For some reason, it stopped after 7 epochs. Maybe there is still something weird going on.

Here is experiment "BPSSynth+BPS on BPS-val"

Epoch 10/100
  1/197 [..............................] - ETA: 2:25 - loss: 3.3874 - key_loss: 0.5963 - degree_1_loss: 0.3263 - degree_2_loss: 0.7435 - quality_loss: 0.5636 - inversion_loss: 0.6117 - root_loss: 0.5461 - key_ac  2/197 [..............................] - ETA: 1:50 - loss: 3.1494 - key_loss: 0.5694 - degree_1_loss: 0.3145 - degree_2_loss: 0.6763 - quality_loss: 0.5336 - inversion_loss: 0.5717 - root_loss: 0.4838 - key_ac  3/197 [..............................] - ETA: 1:48 - loss: 3.1200 - key_loss: 0.5549 - degree_1_loss: 0.3121 - degree_2_loss: 0.6604 - quality_loss: 0.5452 - inversion_loss: 0.5716 - root_loss: 0.4758 - key_ac  4/197 [..............................] - ETA: 1:48 - loss: 3.0660 - key_loss: 0.5333 - degree_1_loss: 0.3063 - degree_2_loss: 0.6478 - quality_loss: 0.5446 - inversion_loss: 0.5687 - root_loss: 0.4653 - key_ac  5/197 [..............................] - ETA: 1:48 - loss: 3.0242 - key_loss: 0.5200 - degree_1_loss: 0.3020 - degree_2_loss: 0.6390 - quality_loss: 0.5435 - inversion_loss: 0.5619 - root_loss: 0.4578 - key_ac  6/197 [..............................] - ETA: 1:47 - loss: 3.0090 - key_loss: 0.5143 - degree_1_loss: 0.3047 - degree_2_loss: 0.6337 - quality_loss: 0.5421 - inversion_loss: 0.5606 - root_loss: 0.4536 - key_ac  7/197 [>.............................] - ETA: 1:46 - loss: 3.0073 - key_loss: 0.5148 - degree_1_loss: 0.3071 - degree_2_loss: 0.6331 - quality_loss: 0.5414 - inversion_loss: 0.5597 - root_loss: 0.4512 - key_ac  8/197 [>.............................] - ETA: 1:46 - loss: 2.9976 - key_loss: 0.5150 - degree_1_loss: 0.3095 - degree_2_loss: 0.6304 - quality_loss: 0.5394 - inversion_loss: 0.5563 - root_loss: 0.4470 - key_ac  9/197 [>.............................] - ETA: 1:46 - loss: 2.9888 - key_loss: 0.5141 - degree_1_loss: 0.3112 - degree_2_loss: 0.6277 - quality_loss: 0.5382 - inversion_loss: 0.5531 - root_loss: 0.4445 - key_ac 10/197 [>.............................] - ETA: 1:46 - loss: 2.9830 - key_loss: 0.5140 - degree_1_loss: 0.3129 - degree_2_loss: 0.6253 - quality_loss: 0.5369 - inversion_loss: 0.5507 - root_loss: 0.4431 - key_ac 11/197 [>.............................] - ETA: 1:45 - loss: 2.9762 - key_loss: 0.5138 - degree_1_loss: 0.3143 - degree_2_loss: 0.6227 - quality_loss: 0.5354 - inversion_loss: 0.5482 - root_loss: 0.4419 - key_ac 12/197 [>.............................] - ETA: 1:45 - loss: 2.9692 - key_loss: 0.5139 - degree_1_loss: 0.3159 - degree_2_loss: 0.6197 - quality_loss: 0.5335 - inversion_loss: 0.5454 - root_loss: 0.4408 - key_ac 13/197 [>.............................] - ETA: 1:44 - loss: 2.9597 - key_loss: 0.5127 - degree_1_loss: 0.3172 - degree_2_loss: 0.6166 - quality_loss: 0.5311 - inversion_loss: 0.5429 - root_loss: 0.4392 - key_ac 14/197 [=>............................] - ETA: 1:44 - loss: 2.9512 - key_loss: 0.5114 - degree_1_loss: 0.3186 - degree_2_loss: 0.6142 - quality_loss: 0.5288 - inversion_loss: 0.5403 - root_loss: 0.4378 - key_ac 15/197 [=>............................] - ETA: 1:43 - loss: 2.9437 - key_loss: 0.5097 - degree_1_loss: 0.3197 - degree_2_loss: 0.6120 - quality_loss: 0.5273 - inversion_loss: 0.5379 - root_loss: 0.4371 - key_ac 16/197 [=>............................] - ETA: 1:42 - loss: 2.9381 - key_loss: 0.5081 - degree_1_loss: 0.3205 - degree_2_loss: 0.6104 - quality_loss: 0.5261 - inversion_loss: 0.5361 - root_loss: 0.4368 - key_ac 17/197 [=>............................] - ETA: 1:42 - loss: 2.9324 - key_loss: 0.5065 - degree_1_loss: 0.3210 - degree_2_loss: 0.6090 - quality_loss: 0.5250 - inversion_loss: 0.5341 - root_loss: 0.4367 - key_ac 18/197 [=>............................] - ETA: 1:41 - loss: 2.9293 - key_loss: 0.5050 - degree_1_loss: 0.3217 - degree_2_loss: 0.6082 - quality_loss: 0.5245 - inversion_loss: 0.5329 - root_loss: 0.4370 - key_ac 19/197 [=>............................] - ETA: 1:41 - loss: 2.9277 - key_loss: 0.5036 - degree_1_loss: 0.3226 - degree_2_loss: 0.6078 - quality_loss: 0.5242 - inversion_loss: 0.5321 - root_loss: 0.4374 - key_ac 20/197 [==>...........................] - ETA: 1:40 - loss: 2.9246 - key_loss: 0.5022 - degree_1_loss: 0.3232 - degree_2_loss: 0.6072 - quality_loss: 0.5236 - inversion_loss: 0.5312 - root_loss: 0.4372 - key_ac 21/197 [==>...........................] - ETA: 1:40 - loss: 2.9226 - key_loss: 0.5007 - degree_1_loss: 0.3240 - degree_2_loss: 0.6068 - quality_loss: 0.5232 - inversion_loss: 0.5308 - root_loss: 0.4372 - key_ac 22/197 [==>...........................] - ETA: 1:39 - loss: 2.9198 - key_loss: 0.4991 - degree_1_loss: 0.3245 - degree_2_loss: 0.6062 - quality_loss: 0.5228 - inversion_loss: 0.5304 - root_loss: 0.4369 - key_ac 23/197 [==>...........................] - ETA: 1:39 - loss: 2.9152 - key_loss: 0.4973 - degree_1_loss: 0.3247 - degree_2_loss: 0.6053 - quality_loss: 0.5221 - inversion_loss: 0.5296 - root_loss: 0.4361 - key_ac 24/197 [==>...........................] - ETA: 1:38 - loss: 2.9118 - key_loss: 0.4958 - degree_1_loss: 0.3250 - degree_2_loss: 0.6048 - quality_loss: 0.5217 - inversion_loss: 0.5289 - root_loss: 0.4357 - key_ac 25/197 [==>...........................] - ETA: 1:38 - loss: 2.9088 - key_loss: 0.4946 - degree_1_loss: 0.3251 - degree_2_loss: 0.6042 - quality_loss: 0.5213 - inversion_loss: 0.5284 - root_loss: 0.4352 - key_ac 26/197 [==>...........................] - ETA: 1:37 - loss: 2.9063 - key_loss: 0.4934 - degree_1_loss: 0.3251 - degree_2_loss: 0.6039 - quality_loss: 0.5212 - inversion_loss: 0.5280 - root_loss: 0.4347 - key_ac 27/197 [===>..........................] - ETA: 1:36 - loss: 2.9041 - key_loss: 0.4923 - degree_1_loss: 0.3251 - degree_2_loss: 0.6036 - quality_loss: 0.5211 - inversion_loss: 0.5277 - root_loss: 0.4343 - key_ac 28/197 [===>..........................] - ETA: 1:36 - loss: 2.9018 - key_loss: 0.4910 - degree_1_loss: 0.3249 - degree_2_loss: 0.6034 - quality_loss: 0.5211 - inversion_loss: 0.5275 - root_loss: 0.4340 - key_ac 29/197 [===>..........................] - ETA: 1:35 - loss: 2.8994 - key_loss: 0.4897 - degree_1_loss: 0.3247 - degree_2_loss: 0.6031 - quality_loss: 0.5210 - inversion_loss: 0.5273 - root_loss: 0.4336 - key_ac 30/197 [===>..........................] - ETA: 1:35 - loss: 2.8971 - key_loss: 0.4883 - degree_1_loss: 0.3244 - degree_2_loss: 0.6029 - quality_loss: 0.5209 - inversion_loss: 0.5272 - root_loss: 0.4333 - key_ac 31/197 [===>..........................] - ETA: 1:34 - loss: 2.8954 - key_loss: 0.4870 - degree_1_loss: 0.3242 - degree_2_loss: 0.6027 - quality_loss: 0.5210 - inversion_loss: 0.5274 - root_loss: 0.4332 - key_ac 32/197 [===>..........................] - ETA: 1:34 - loss: 2.8936 - key_loss: 0.4859 - degree_1_loss: 0.3238 - degree_2_loss: 0.6025 - quality_loss: 0.5211 - inversion_loss: 0.5274 - root_loss: 0.4329 - key_ac 33/197 [====>.........................] - ETA: 1:33 - loss: 2.8917 - key_loss: 0.4847 - degree_1_loss: 0.3235 - degree_2_loss: 0.6024 - quality_loss: 0.5211 - inversion_loss: 0.5274 - root_loss: 0.4327 - key_ac 34/197 [====>.........................] - ETA: 1:33 - loss: 2.8901 - key_loss: 0.4837 - degree_1_loss: 0.3232 - degree_2_loss: 0.6023 - quality_loss: 0.5212 - inversion_loss: 0.5273 - root_loss: 0.4325 - key_ac 35/197 [====>.........................] - ETA: 1:32 - loss: 2.8888 - key_loss: 0.4828 - degree_1_loss: 0.3229 - degree_2_loss: 0.6022 - quality_loss: 0.5213 - inversion_loss: 0.5273 - root_loss: 0.4323 - key_ac 36/197 [====>.........................] - ETA: 1:31 - loss: 2.8877 - key_loss: 0.4818 - degree_1_loss: 0.3225 - degree_2_loss: 0.6021 - quality_loss: 0.5215 - inversion_loss: 0.5274 - root_loss: 0.4323 - key_ac 37/197 [====>.........................] - ETA: 1:31 - loss: 2.8871 - key_loss: 0.4811 - degree_1_loss: 0.3223 - degree_2_loss: 0.6022 - quality_loss: 0.5217 - inversion_loss: 0.5275 - root_loss: 0.4323 - key_ac 38/197 [====>.........................] - ETA: 1:30 - loss: 2.8868 - key_loss: 0.4805 - degree_1_loss: 0.3221 - degree_2_loss: 0.6022 - quality_loss: 0.5221 - inversion_loss: 0.5275 - root_loss: 0.4324 - key_ac 39/197 [====>.........................] - ETA: 1:30 - loss: 2.8864 - key_loss: 0.4799 - degree_1_loss: 0.3220 - degree_2_loss: 0.6022 - quality_loss: 0.5223 - inversion_loss: 0.5275 - root_loss: 0.4325 - key_ac 40/197 [=====>........................] - ETA: 1:29 - loss: 2.8863 - key_loss: 0.4794 - degree_1_loss: 0.3219 - degree_2_loss: 0.6023 - quality_loss: 0.5226 - inversion_loss: 0.5276 - root_loss: 0.4326 - key_ac 41/197 [=====>........................] - ETA: 1:29 - loss: 2.8866 - key_loss: 0.4790 - degree_1_loss: 0.3219 - degree_2_loss: 0.6024 - quality_loss: 0.5229 - inversion_loss: 0.5277 - root_loss: 0.4327 - key_ac 42/197 [=====>........................] - ETA: 1:28 - loss: 2.8868 - key_loss: 0.4784 - degree_1_loss: 0.3218 - degree_2_loss: 0.6025 - quality_loss: 0.5233 - inversion_loss: 0.5279 - root_loss: 0.4329 - key_ac 43/197 [=====>........................] - ETA: 1:28 - loss: 2.8869 - key_loss: 0.4779 - degree_1_loss: 0.3217 - degree_2_loss: 0.6026 - quality_loss: 0.5236 - inversion_loss: 0.5280 - root_loss: 0.4331 - key_ac 44/197 [=====>........................] - ETA: 1:27 - loss: 2.8868 - key_loss: 0.4775 - degree_1_loss: 0.3216 - degree_2_loss: 0.6026 - quality_loss: 0.5239 - inversion_loss: 0.5280 - root_loss: 0.4331 - key_ac 45/197 [=====>........................] - ETA: 1:27 - loss: 2.8870 - key_loss: 0.4770 - degree_1_loss: 0.3216 - degree_2_loss: 0.6027 - quality_loss: 0.5242 - inversion_loss: 0.5282 - root_loss: 0.4333 - key_ac 46/197 [======>.......................] - ETA: 1:26 - loss: 2.8870 - key_loss: 0.4764 - degree_1_loss: 0.3215 - degree_2_loss: 0.6028 - quality_loss: 0.5245 - inversion_loss: 0.5283 - root_loss: 0.4335 - key_ac 47/197 [======>.......................] - ETA: 1:25 - loss: 2.8872 - key_loss: 0.4759 - degree_1_loss: 0.3215 - degree_2_loss: 0.6030 - quality_loss: 0.5248 - inversion_loss: 0.5284 - root_loss: 0.4337 - key_ac 48/197 [======>.......................] - ETA: 1:25 - loss: 2.8876 - key_loss: 0.4755 - degree_1_loss: 0.3214 - degree_2_loss: 0.6031 - quality_loss: 0.5251 - inversion_loss: 0.5286 - root_loss: 0.4339 - key_ac 49/197 [======>.......................] - ETA: 1:24 - loss: 2.8878 - key_loss: 0.4750 - degree_1_loss: 0.3214 - degree_2_loss: 0.6033 - quality_loss: 0.5253 - inversion_loss: 0.5287 - root_loss: 0.4341 - key_ac 50/197 [======>.......................] - ETA: 1:24 - loss: 2.8879 - key_loss: 0.4746 - degree_1_loss: 0.3213 - degree_2_loss: 0.6034 - quality_loss: 0.5255 - inversion_loss: 0.5289 - root_loss: 0.4343 - key_ac 51/197 [======>.......................] - ETA: 1:23 - loss: 2.8880 - key_loss: 0.4741 - degree_1_loss: 0.3213 - degree_2_loss: 0.6034 - quality_loss: 0.5257 - inversion_loss: 0.5291 - root_loss: 0.4344 - key_ac 52/197 [======>.......................] - ETA: 1:22 - loss: 2.8881 - key_loss: 0.4736 - degree_1_loss: 0.3213 - degree_2_loss: 0.6035 - quality_loss: 0.5259 - inversion_loss: 0.5292 - root_loss: 0.4346 - key_ac 53/197 [=======>......................] - ETA: 1:22 - loss: 2.8883 - key_loss: 0.4731 - degree_1_loss: 0.3213 - degree_2_loss: 0.6037 - quality_loss: 0.5261 - inversion_loss: 0.5294 - root_loss: 0.4347 - key_ac 54/197 [=======>......................] - ETA: 1:21 - loss: 2.8885 - key_loss: 0.4726 - degree_1_loss: 0.3212 - degree_2_loss: 0.6038 - quality_loss: 0.5264 - inversion_loss: 0.5296 - root_loss: 0.4349 - key_ac 55/197 [=======>......................] - ETA: 1:21 - loss: 2.8890 - key_loss: 0.4722 - degree_1_loss: 0.3213 - degree_2_loss: 0.6040 - quality_loss: 0.5266 - inversion_loss: 0.5299 - root_loss: 0.4351 - key_ac 56/197 [=======>......................] - ETA: 1:20 - loss: 2.8894 - key_loss: 0.4718 - degree_1_loss: 0.3213 - degree_2_loss: 0.6042 - quality_loss: 0.5269 - inversion_loss: 0.5301 - root_loss: 0.4352 - key_ac 57/197 [=======>......................] - ETA: 1:20 - loss: 2.8898 - key_loss: 0.4714 - degree_1_loss: 0.3213 - degree_2_loss: 0.6043 - quality_loss: 0.5271 - inversion_loss: 0.5304 - root_loss: 0.4354 - key_ac 58/197 [=======>......................] - ETA: 1:19 - loss: 2.8903 - key_loss: 0.4710 - degree_1_loss: 0.3213 - degree_2_loss: 0.6044 - quality_loss: 0.5273 - inversion_loss: 0.5306 - root_loss: 0.4356 - key_ac 59/197 [=======>......................] - ETA: 1:18 - loss: 2.8909 - key_loss: 0.4706 - degree_1_loss: 0.3214 - degree_2_loss: 0.6046 - quality_loss: 0.5276 - inversion_loss: 0.5309 - root_loss: 0.4358 - key_ac 60/197 [========>.....................] - ETA: 1:18 - loss: 2.8916 - key_loss: 0.4704 - degree_1_loss: 0.3214 - degree_2_loss: 0.6048 - quality_loss: 0.5278 - inversion_loss: 0.5312 - root_loss: 0.4360 - key_ac 61/197 [========>.....................] - ETA: 1:17 - loss: 2.8923 - key_loss: 0.4701 - degree_1_loss: 0.3215 - degree_2_loss: 0.6049 - quality_loss: 0.5280 - inversion_loss: 0.5315 - root_loss: 0.4363 - key_ac 62/197 [========>.....................] - ETA: 1:17 - loss: 2.8930 - key_loss: 0.4699 - degree_1_loss: 0.3216 - degree_2_loss: 0.6051 - quality_loss: 0.5283 - inversion_loss: 0.5317 - root_loss: 0.4365 - key_ac 63/197 [========>.....................] - ETA: 1:16 - loss: 2.8936 - key_loss: 0.4696 - degree_1_loss: 0.3216 - degree_2_loss: 0.6052 - quality_loss: 0.5285 - inversion_loss: 0.5320 - root_loss: 0.4366 - key_ac 64/197 [========>.....................] - ETA: 1:16 - loss: 2.8940 - key_loss: 0.4694 - degree_1_loss: 0.3217 - degree_2_loss: 0.6053 - quality_loss: 0.5286 - inversion_loss: 0.5322 - root_loss: 0.4368 - key_ac 65/197 [========>.....................] - ETA: 1:15 - loss: 2.8943 - key_loss: 0.4692 - degree_1_loss: 0.3217 - degree_2_loss: 0.6054 - quality_loss: 0.5287 - inversion_loss: 0.5324 - root_loss: 0.4369 - key_ac 66/197 [=========>....................] - ETA: 1:14 - loss: 2.8946 - key_loss: 0.4690 - degree_1_loss: 0.3218 - degree_2_loss: 0.6055 - quality_loss: 0.5288 - inversion_loss: 0.5325 - root_loss: 0.4370 - key_ac 67/197 [=========>....................] - ETA: 1:14 - loss: 2.8948 - key_loss: 0.4688 - degree_1_loss: 0.3218 - degree_2_loss: 0.6056 - quality_loss: 0.5289 - inversion_loss: 0.5327 - root_loss: 0.4371 - key_ac 68/197 [=========>....................] - ETA: 1:13 - loss: 2.8950 - key_loss: 0.4685 - degree_1_loss: 0.3218 - degree_2_loss: 0.6056 - quality_loss: 0.5290 - inversion_loss: 0.5328 - root_loss: 0.4371 - key_ac 69/197 [=========>....................] - ETA: 1:13 - loss: 2.8951 - key_loss: 0.4683 - degree_1_loss: 0.3218 - degree_2_loss: 0.6057 - quality_loss: 0.5291 - inversion_loss: 0.5329 - root_loss: 0.4372 - key_ac 70/197 [=========>....................] - ETA: 1:12 - loss: 2.8950 - key_loss: 0.4680 - degree_1_loss: 0.3218 - degree_2_loss: 0.6058 - quality_loss: 0.5292 - inversion_loss: 0.5330 - root_loss: 0.4373 - key_ac 71/197 [=========>....................] - ETA: 1:12 - loss: 2.8949 - key_loss: 0.4677 - degree_1_loss: 0.3217 - degree_2_loss: 0.6058 - quality_loss: 0.5292 - inversion_loss: 0.5331 - root_loss: 0.4373 - key_ac 72/197 [=========>....................] - ETA: 1:11 - loss: 2.8948 - key_loss: 0.4675 - degree_1_loss: 0.3217 - degree_2_loss: 0.6058 - quality_loss: 0.5292 - inversion_loss: 0.5332 - root_loss: 0.4373 - key_ac 73/197 [==========>...................] - ETA: 1:10 - loss: 2.8948 - key_loss: 0.4673 - degree_1_loss: 0.3217 - degree_2_loss: 0.6059 - quality_loss: 0.5293 - inversion_loss: 0.5333 - root_loss: 0.4374 - key_ac 74/197 [==========>...................] - ETA: 1:10 - loss: 2.8949 - key_loss: 0.4671 - degree_1_loss: 0.3217 - degree_2_loss: 0.6059 - quality_loss: 0.5293 - inversion_loss: 0.5334 - root_loss: 0.4374 - key_ac 75/197 [==========>...................] - ETA: 1:09 - loss: 2.8950 - key_loss: 0.4670 - degree_1_loss: 0.3217 - degree_2_loss: 0.6059 - quality_loss: 0.5294 - inversion_loss: 0.5335 - root_loss: 0.4374 - key_ac 76/197 [==========>...................] - ETA: 1:09 - loss: 2.8951 - key_loss: 0.4668 - degree_1_loss: 0.3218 - degree_2_loss: 0.6060 - quality_loss: 0.5294 - inversion_loss: 0.5336 - root_loss: 0.4375 - key_ac 77/197 [==========>...................] - ETA: 1:08 - loss: 2.8953 - key_loss: 0.4668 - degree_1_loss: 0.3218 - degree_2_loss: 0.6061 - quality_loss: 0.5295 - inversion_loss: 0.5337 - root_loss: 0.4376 - key_ac 78/197 [==========>...................] - ETA: 1:08 - loss: 2.8956 - key_loss: 0.4667 - degree_1_loss: 0.3218 - degree_2_loss: 0.6062 - quality_loss: 0.5295 - inversion_loss: 0.5338 - root_loss: 0.4376 - key_ac 79/197 [===========>..................] - ETA: 1:07 - loss: 2.8959 - key_loss: 0.4666 - degree_1_loss: 0.3219 - degree_2_loss: 0.6062 - quality_loss: 0.5296 - inversion_loss: 0.5339 - root_loss: 0.4377 - key_ac 80/197 [===========>..................] - ETA: 1:06 - loss: 2.8962 - key_loss: 0.4665 - degree_1_loss: 0.3219 - degree_2_loss: 0.6063 - quality_loss: 0.5296 - inversion_loss: 0.5340 - root_loss: 0.4378 - key_ac 81/197 [===========>..................] - ETA: 1:06 - loss: 2.8965 - key_loss: 0.4665 - degree_1_loss: 0.3220 - degree_2_loss: 0.6064 - quality_loss: 0.5297 - inversion_loss: 0.5341 - root_loss: 0.4378 - key_ac 82/197 [===========>..................] - ETA: 1:05 - loss: 2.8967 - key_loss: 0.4664 - degree_1_loss: 0.3220 - degree_2_loss: 0.6065 - quality_loss: 0.5297 - inversion_loss: 0.5342 - root_loss: 0.4379 - key_ac 83/197 [===========>..................] - ETA: 1:05 - loss: 2.8970 - key_loss: 0.4663 - degree_1_loss: 0.3220 - degree_2_loss: 0.6066 - quality_loss: 0.5298 - inversion_loss: 0.5343 - root_loss: 0.4379 - key_ac 84/197 [===========>..................] - ETA: 1:04 - loss: 2.8971 - key_loss: 0.4662 - degree_1_loss: 0.3220 - degree_2_loss: 0.6067 - quality_loss: 0.5299 - inversion_loss: 0.5344 - root_loss: 0.4379 - key_ac 85/197 [===========>..................] - ETA: 1:04 - loss: 2.8973 - key_loss: 0.4662 - degree_1_loss: 0.3220 - degree_2_loss: 0.6068 - quality_loss: 0.5299 - inversion_loss: 0.5344 - root_loss: 0.4380 - key_ac 86/197 [============>.................] - ETA: 1:03 - loss: 2.8975 - key_loss: 0.4661 - degree_1_loss: 0.3221 - degree_2_loss: 0.6068 - quality_loss: 0.5300 - inversion_loss: 0.5345 - root_loss: 0.4380 - key_ac 87/197 [============>.................] - ETA: 1:03 - loss: 2.8977 - key_loss: 0.4661 - degree_1_loss: 0.3221 - degree_2_loss: 0.6069 - quality_loss: 0.5300 - inversion_loss: 0.5346 - root_loss: 0.4380 - key_ac 88/197 [============>.................] - ETA: 1:02 - loss: 2.8978 - key_loss: 0.4660 - degree_1_loss: 0.3221 - degree_2_loss: 0.6070 - quality_loss: 0.5301 - inversion_loss: 0.5347 - root_loss: 0.4380 - key_ac 89/197 [============>.................] - ETA: 1:01 - loss: 2.8979 - key_loss: 0.4659 - degree_1_loss: 0.3222 - degree_2_loss: 0.6070 - quality_loss: 0.5301 - inversion_loss: 0.5347 - root_loss: 0.4380 - key_ac 90/197 [============>.................] - ETA: 1:01 - loss: 2.8981 - key_loss: 0.4659 - degree_1_loss: 0.3222 - degree_2_loss: 0.6071 - quality_loss: 0.5301 - inversion_loss: 0.5348 - root_loss: 0.4380 - key_ac 91/197 [============>.................] - ETA: 1:00 - loss: 2.8982 - key_loss: 0.4658 - degree_1_loss: 0.3222 - degree_2_loss: 0.6071 - quality_loss: 0.5302 - inversion_loss: 0.5348 - root_loss: 0.4380 - key_ac 92/197 [=============>................] - ETA: 1:00 - loss: 2.8982 - key_loss: 0.4657 - degree_1_loss: 0.3222 - degree_2_loss: 0.6072 - quality_loss: 0.5302 - inversion_loss: 0.5349 - root_loss: 0.4380 - key_ac 93/197 [=============>................] - ETA: 59s - loss: 2.8981 - key_loss: 0.4656 - degree_1_loss: 0.3222 - degree_2_loss: 0.6072 - quality_loss: 0.5302 - inversion_loss: 0.5349 - root_loss: 0.4380 - key_acc 94/197 [=============>................] - ETA: 58s - loss: 2.8980 - key_loss: 0.4656 - degree_1_loss: 0.3222 - degree_2_loss: 0.6072 - quality_loss: 0.5302 - inversion_loss: 0.5349 - root_loss: 0.4380 - key_acc 95/197 [=============>................] - ETA: 58s - loss: 2.8978 - key_loss: 0.4655 - degree_1_loss: 0.3222 - degree_2_loss: 0.6072 - quality_loss: 0.5301 - inversion_loss: 0.5349 - root_loss: 0.4380 - key_acc 96/197 [=============>................] - ETA: 57s - loss: 2.8977 - key_loss: 0.4654 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5301 - inversion_loss: 0.5349 - root_loss: 0.4379 - key_acc 97/197 [=============>................] - ETA: 57s - loss: 2.8976 - key_loss: 0.4653 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5301 - inversion_loss: 0.5349 - root_loss: 0.4379 - key_acc 98/197 [=============>................] - ETA: 56s - loss: 2.8975 - key_loss: 0.4653 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5301 - inversion_loss: 0.5349 - root_loss: 0.4379 - key_acc 99/197 [==============>...............] - ETA: 56s - loss: 2.8974 - key_loss: 0.4652 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5300 - inversion_loss: 0.5349 - root_loss: 0.4379 - key_acc100/197 [==============>...............] - ETA: 55s - loss: 2.8973 - key_loss: 0.4652 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5300 - inversion_loss: 0.5349 - root_loss: 0.4379 - key_acc101/197 [==============>...............] - ETA: 54s - loss: 2.8972 - key_loss: 0.4651 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5300 - inversion_loss: 0.5349 - root_loss: 0.4378 - key_acc102/197 [==============>...............] - ETA: 54s - loss: 2.8970 - key_loss: 0.4650 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5299 - inversion_loss: 0.5349 - root_loss: 0.4378 - key_acc103/197 [==============>...............] - ETA: 53s - loss: 2.8968 - key_loss: 0.4650 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5299 - inversion_loss: 0.5349 - root_loss: 0.4378 - key_acc104/197 [==============>...............] - ETA: 53s - loss: 2.8967 - key_loss: 0.4649 - degree_1_loss: 0.3221 - degree_2_loss: 0.6072 - quality_loss: 0.5298 - inversion_loss: 0.5350 - root_loss: 0.4378 - key_acc105/197 [==============>...............] - ETA: 52s - loss: 2.8966 - key_loss: 0.4649 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5298 - inversion_loss: 0.5350 - root_loss: 0.4377 - key_acc106/197 [===============>..............] - ETA: 52s - loss: 2.8965 - key_loss: 0.4648 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5297 - inversion_loss: 0.5350 - root_loss: 0.4377 - key_acc107/197 [===============>..............] - ETA: 51s - loss: 2.8964 - key_loss: 0.4648 - degree_1_loss: 0.3220 - degree_2_loss: 0.6071 - quality_loss: 0.5297 - inversion_loss: 0.5350 - root_loss: 0.4377 - key_acc108/197 [===============>..............] - ETA: 50s - loss: 2.8962 - key_loss: 0.4647 - degree_1_loss: 0.3220 - degree_2_loss: 0.6071 - quality_loss: 0.5296 - inversion_loss: 0.5350 - root_loss: 0.4377 - key_acc109/197 [===============>..............] - ETA: 50s - loss: 2.8961 - key_loss: 0.4647 - degree_1_loss: 0.3220 - degree_2_loss: 0.6071 - quality_loss: 0.5296 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc110/197 [===============>..............] - ETA: 49s - loss: 2.8960 - key_loss: 0.4646 - degree_1_loss: 0.3220 - degree_2_loss: 0.6071 - quality_loss: 0.5295 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc111/197 [===============>..............] - ETA: 49s - loss: 2.8959 - key_loss: 0.4646 - degree_1_loss: 0.3220 - degree_2_loss: 0.6071 - quality_loss: 0.5295 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc112/197 [================>.............] - ETA: 48s - loss: 2.8958 - key_loss: 0.4645 - degree_1_loss: 0.3220 - degree_2_loss: 0.6071 - quality_loss: 0.5295 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc113/197 [================>.............] - ETA: 48s - loss: 2.8957 - key_loss: 0.4645 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5294 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc114/197 [================>.............] - ETA: 47s - loss: 2.8957 - key_loss: 0.4645 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5294 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc115/197 [================>.............] - ETA: 46s - loss: 2.8956 - key_loss: 0.4644 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5294 - inversion_loss: 0.5351 - root_loss: 0.4376 - key_acc116/197 [================>.............] - ETA: 46s - loss: 2.8956 - key_loss: 0.4644 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5293 - inversion_loss: 0.5351 - root_loss: 0.4375 - key_acc117/197 [================>.............] - ETA: 45s - loss: 2.8956 - key_loss: 0.4644 - degree_1_loss: 0.3220 - degree_2_loss: 0.6072 - quality_loss: 0.5293 - inversion_loss: 0.5351 - root_loss: 0.4375 - key_acc118/197 [================>.............] - ETA: 45s - loss: 2.8955 - key_loss: 0.4643 - degree_1_loss: 0.3220 - degree_2_loss: 0.6073 - quality_loss: 0.5293 - inversion_loss: 0.5351 - root_loss: 0.4375 - key_acc119/197 [=================>............] - ETA: 44s - loss: 2.8955 - key_loss: 0.4643 - degree_1_loss: 0.3220 - degree_2_loss: 0.6073 - quality_loss: 0.5292 - inversion_loss: 0.5351 - root_loss: 0.4375 - key_acc120/197 [=================>............] - ETA: 44s - loss: 2.8955 - key_loss: 0.4643 - degree_1_loss: 0.3220 - degree_2_loss: 0.6073 - quality_loss: 0.5292 - inversion_loss: 0.5351 - root_loss: 0.4375 - key_acc121/197 [=================>............] - ETA: 43s - loss: 2.8955 - key_loss: 0.4643 - degree_1_loss: 0.3220 - degree_2_loss: 0.6074 - quality_loss: 0.5292 - inversion_loss: 0.5351 - root_loss: 0.4375 - key_acc122/197 [=================>............] - ETA: 42s - loss: 2.8955 - key_loss: 0.4643 - degree_1_loss: 0.3221 - degree_2_loss: 0.6074 - quality_loss: 0.5292 - inversion_loss: 0.5352 - root_loss: 0.4375 - key_acc123/197 [=================>............] - ETA: 42s - loss: 2.8956 - key_loss: 0.4643 - degree_1_loss: 0.3221 - degree_2_loss: 0.6074 - quality_loss: 0.5292 - inversion_loss: 0.5352 - root_loss: 0.4375 - key_acc124/197 [=================>............] - ETA: 41s - loss: 2.8957 - key_loss: 0.4642 - degree_1_loss: 0.3221 - degree_2_loss: 0.6075 - quality_loss: 0.5292 - inversion_loss: 0.5352 - root_loss: 0.4375 - key_acc125/197 [==================>...........] - ETA: 41s - loss: 2.8957 - key_loss: 0.4642 - degree_1_loss: 0.3221 - degree_2_loss: 0.6075 - quality_loss: 0.5292 - inversion_loss: 0.5352 - root_loss: 0.4375 - key_acc126/197 [==================>...........] - ETA: 40s - loss: 2.8957 - key_loss: 0.4642 - degree_1_loss: 0.3221 - degree_2_loss: 0.6075 - quality_loss: 0.5291 - inversion_loss: 0.5352 - root_loss: 0.4374 - key_acc127/197 [==================>...........] - ETA: 39s - loss: 2.8958 - key_loss: 0.4642 - degree_1_loss: 0.3222 - degree_2_loss: 0.6076 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4374 - key_acc128/197 [==================>...........] - ETA: 39s - loss: 2.8958 - key_loss: 0.4642 - degree_1_loss: 0.3222 - degree_2_loss: 0.6076 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4374 - key_acc129/197 [==================>...........] - ETA: 38s - loss: 2.8959 - key_loss: 0.4642 - degree_1_loss: 0.3222 - degree_2_loss: 0.6076 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4374 - key_acc130/197 [==================>...........] - ETA: 38s - loss: 2.8959 - key_loss: 0.4642 - degree_1_loss: 0.3222 - degree_2_loss: 0.6077 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4374 - key_acc131/197 [==================>...........] - ETA: 37s - loss: 2.8959 - key_loss: 0.4642 - degree_1_loss: 0.3222 - degree_2_loss: 0.6077 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4374 - key_acc132/197 [===================>..........] - ETA: 37s - loss: 2.8959 - key_loss: 0.4642 - degree_1_loss: 0.3223 - degree_2_loss: 0.6077 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4373 - key_acc133/197 [===================>..........] - ETA: 36s - loss: 2.8959 - key_loss: 0.4642 - degree_1_loss: 0.3223 - degree_2_loss: 0.6077 - quality_loss: 0.5291 - inversion_loss: 0.5353 - root_loss: 0.4373 - key_acc134/197 [===================>..........] - ETA: 35s - loss: 2.8960 - key_loss: 0.4642 - degree_1_loss: 0.3223 - degree_2_loss: 0.6078 - quality_loss: 0.5291 - inversion_loss: 0.5354 - root_loss: 0.4373 - key_acc135/197 [===================>..........] - ETA: 35s - loss: 2.8960 - key_loss: 0.4642 - degree_1_loss: 0.3223 - degree_2_loss: 0.6078 - quality_loss: 0.5291 - inversion_loss: 0.5354 - root_loss: 0.4373 - key_acc136/197 [===================>..........] - ETA: 34s - loss: 2.8961 - key_loss: 0.4642 - degree_1_loss: 0.3224 - degree_2_loss: 0.6078 - quality_loss: 0.5291 - inversion_loss: 0.5354 - root_loss: 0.4373 - key_acc137/197 [===================>..........] - ETA: 34s - loss: 2.8962 - key_loss: 0.4642 - degree_1_loss: 0.3224 - degree_2_loss: 0.6079 - quality_loss: 0.5291 - inversion_loss: 0.5354 - root_loss: 0.4373 - key_acc138/197 [====================>.........] - ETA: 33s - loss: 2.8963 - key_loss: 0.4642 - degree_1_loss: 0.3224 - degree_2_loss: 0.6079 - quality_loss: 0.5291 - inversion_loss: 0.5355 - root_loss: 0.4373 - key_acc139/197 [====================>.........] - ETA: 33s - loss: 2.8964 - key_loss: 0.4642 - degree_1_loss: 0.3224 - degree_2_loss: 0.6079 - quality_loss: 0.5291 - inversion_loss: 0.5355 - root_loss: 0.4373 - key_acc140/197 [====================>.........] - ETA: 32s - loss: 2.8965 - key_loss: 0.4642 - degree_1_loss: 0.3225 - degree_2_loss: 0.6080 - quality_loss: 0.5291 - inversion_loss: 0.5355 - root_loss: 0.4373 - key_acc141/197 [====================>.........] - ETA: 31s - loss: 2.8967 - key_loss: 0.4642 - degree_1_loss: 0.3225 - degree_2_loss: 0.6080 - quality_loss: 0.5291 - inversion_loss: 0.5356 - root_loss: 0.4373 - key_acc142/197 [====================>.........] - ETA: 31s - loss: 2.8968 - key_loss: 0.4642 - degree_1_loss: 0.3225 - degree_2_loss: 0.6080 - quality_loss: 0.5291 - inversion_loss: 0.5356 - root_loss: 0.4373 - key_acc143/197 [====================>.........] - ETA: 30s - loss: 2.8969 - key_loss: 0.4642 - degree_1_loss: 0.3226 - degree_2_loss: 0.6081 - quality_loss: 0.5291 - inversion_loss: 0.5356 - root_loss: 0.4373 - key_acc144/197 [====================>.........] - ETA: 30s - loss: 2.8971 - key_loss: 0.4642 - degree_1_loss: 0.3226 - degree_2_loss: 0.6081 - quality_loss: 0.5291 - inversion_loss: 0.5357 - root_loss: 0.4373 - key_acc145/197 [=====================>........] - ETA: 29s - loss: 2.8972 - key_loss: 0.4642 - degree_1_loss: 0.3226 - degree_2_loss: 0.6082 - quality_loss: 0.5292 - inversion_loss: 0.5357 - root_loss: 0.4373 - key_acc146/197 [=====================>........] - ETA: 29s - loss: 2.8973 - key_loss: 0.4642 - degree_1_loss: 0.3227 - degree_2_loss: 0.6082 - quality_loss: 0.5292 - inversion_loss: 0.5358 - root_loss: 0.4373 - key_acc147/197 [=====================>........] - ETA: 28s - loss: 2.8974 - key_loss: 0.4642 - degree_1_loss: 0.3227 - degree_2_loss: 0.6082 - quality_loss: 0.5292 - inversion_loss: 0.5358 - root_loss: 0.4373 - key_acc148/197 [=====================>........] - ETA: 27s - loss: 2.8975 - key_loss: 0.4642 - degree_1_loss: 0.3227 - degree_2_loss: 0.6083 - quality_loss: 0.5292 - inversion_loss: 0.5358 - root_loss: 0.4373 - key_acc149/197 [=====================>........] - ETA: 27s - loss: 2.8976 - key_loss: 0.4642 - degree_1_loss: 0.3228 - degree_2_loss: 0.6083 - quality_loss: 0.5292 - inversion_loss: 0.5359 - root_loss: 0.4373 - key_acc150/197 [=====================>........] - ETA: 26s - loss: 2.8976 - key_loss: 0.4642 - degree_1_loss: 0.3228 - degree_2_loss: 0.6083 - quality_loss: 0.5292 - inversion_loss: 0.5359 - root_loss: 0.4373 - key_acc151/197 [=====================>........] - ETA: 26s - loss: 2.8977 - key_loss: 0.4642 - degree_1_loss: 0.3228 - degree_2_loss: 0.6083 - quality_loss: 0.5292 - inversion_loss: 0.5359 - root_loss: 0.4372 - key_acc152/197 [======================>.......] - ETA: 25s - loss: 2.8977 - key_loss: 0.4642 - degree_1_loss: 0.3228 - degree_2_loss: 0.6084 - quality_loss: 0.5292 - inversion_loss: 0.5359 - root_loss: 0.4372 - key_acc153/197 [======================>.......] - ETA: 25s - loss: 2.8978 - key_loss: 0.4642 - degree_1_loss: 0.3229 - degree_2_loss: 0.6084 - quality_loss: 0.5292 - inversion_loss: 0.5360 - root_loss: 0.4372 - key_acc154/197 [======================>.......] - ETA: 24s - loss: 2.8978 - key_loss: 0.4641 - degree_1_loss: 0.3229 - degree_2_loss: 0.6084 - quality_loss: 0.5292 - inversion_loss: 0.5360 - root_loss: 0.4372 - key_acc155/197 [======================>.......] - ETA: 23s - loss: 2.8978 - key_loss: 0.4641 - degree_1_loss: 0.3229 - degree_2_loss: 0.6084 - quality_loss: 0.5292 - inversion_loss: 0.5360 - root_loss: 0.4372 - key_acc156/197 [======================>.......] - ETA: 23s - loss: 2.8979 - key_loss: 0.4641 - degree_1_loss: 0.3229 - degree_2_loss: 0.6085 - quality_loss: 0.5292 - inversion_loss: 0.5361 - root_loss: 0.4372 - key_acc157/197 [======================>.......] - ETA: 22s - loss: 2.8979 - key_loss: 0.4641 - degree_1_loss: 0.3230 - degree_2_loss: 0.6085 - quality_loss: 0.5292 - inversion_loss: 0.5361 - root_loss: 0.4372 - key_acc158/197 [=======================>......] - ETA: 22s - loss: 2.8980 - key_loss: 0.4641 - degree_1_loss: 0.3230 - degree_2_loss: 0.6085 - quality_loss: 0.5292 - inversion_loss: 0.5361 - root_loss: 0.4371 - key_acc159/197 [=======================>......] - ETA: 21s - loss: 2.8981 - key_loss: 0.4640 - degree_1_loss: 0.3230 - degree_2_loss: 0.6085 - quality_loss: 0.5292 - inversion_loss: 0.5362 - root_loss: 0.4371 - key_acc160/197 [=======================>......] - ETA: 21s - loss: 2.8982 - key_loss: 0.4640 - degree_1_loss: 0.3230 - degree_2_loss: 0.6086 - quality_loss: 0.5292 - inversion_loss: 0.5362 - root_loss: 0.4371 - key_acc161/197 [=======================>......] - ETA: 20s - loss: 2.8983 - key_loss: 0.4640 - degree_1_loss: 0.3231 - degree_2_loss: 0.6086 - quality_loss: 0.5292 - inversion_loss: 0.5362 - root_loss: 0.4371 - key_acc162/197 [=======================>......] - ETA: 20s - loss: 2.8984 - key_loss: 0.4640 - degree_1_loss: 0.3231 - degree_2_loss: 0.6086 - quality_loss: 0.5292 - inversion_loss: 0.5363 - root_loss: 0.4371 - key_acc163/197 [=======================>......] - ETA: 19s - loss: 2.8986 - key_loss: 0.4640 - degree_1_loss: 0.3231 - degree_2_loss: 0.6087 - quality_loss: 0.5292 - inversion_loss: 0.5363 - root_loss: 0.4371 - key_acc164/197 [=======================>......] - ETA: 18s - loss: 2.8987 - key_loss: 0.4640 - degree_1_loss: 0.3232 - degree_2_loss: 0.6087 - quality_loss: 0.5292 - inversion_loss: 0.5364 - root_loss: 0.4371 - key_acc165/197 [========================>.....] - ETA: 18s - loss: 2.8988 - key_loss: 0.4640 - degree_1_loss: 0.3232 - degree_2_loss: 0.6088 - quality_loss: 0.5293 - inversion_loss: 0.5364 - root_loss: 0.4371 - key_acc166/197 [========================>.....] - ETA: 17s - loss: 2.8989 - key_loss: 0.4640 - degree_1_loss: 0.3232 - degree_2_loss: 0.6088 - quality_loss: 0.5293 - inversion_loss: 0.5365 - root_loss: 0.4371 - key_acc167/197 [========================>.....] - ETA: 17s - loss: 2.8990 - key_loss: 0.4640 - degree_1_loss: 0.3232 - degree_2_loss: 0.6088 - quality_loss: 0.5293 - inversion_loss: 0.5365 - root_loss: 0.4371 - key_acc168/197 [========================>.....] - ETA: 16s - loss: 2.8991 - key_loss: 0.4640 - degree_1_loss: 0.3233 - degree_2_loss: 0.6089 - quality_loss: 0.5293 - inversion_loss: 0.5366 - root_loss: 0.4371 - key_acc169/197 [========================>.....] - ETA: 15s - loss: 2.8992 - key_loss: 0.4640 - degree_1_loss: 0.3233 - degree_2_loss: 0.6089 - quality_loss: 0.5293 - inversion_loss: 0.5366 - root_loss: 0.4371 - key_acc170/197 [========================>.....] - ETA: 15s - loss: 2.8994 - key_loss: 0.4640 - degree_1_loss: 0.3233 - degree_2_loss: 0.6089 - quality_loss: 0.5294 - inversion_loss: 0.5367 - root_loss: 0.4371 - key_acc171/197 [=========================>....] - ETA: 14s - loss: 2.8995 - key_loss: 0.4640 - degree_1_loss: 0.3233 - degree_2_loss: 0.6090 - quality_loss: 0.5294 - inversion_loss: 0.5367 - root_loss: 0.4371 - key_acc172/197 [=========================>....] - ETA: 14s - loss: 2.8996 - key_loss: 0.4640 - degree_1_loss: 0.3233 - degree_2_loss: 0.6090 - quality_loss: 0.5294 - inversion_loss: 0.5367 - root_loss: 0.4371 - key_acc173/197 [=========================>....] - ETA: 13s - loss: 2.8997 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6091 - quality_loss: 0.5294 - inversion_loss: 0.5368 - root_loss: 0.4371 - key_acc174/197 [=========================>....] - ETA: 13s - loss: 2.8998 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6091 - quality_loss: 0.5294 - inversion_loss: 0.5368 - root_loss: 0.4371 - key_acc175/197 [=========================>....] - ETA: 12s - loss: 2.9000 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6092 - quality_loss: 0.5295 - inversion_loss: 0.5368 - root_loss: 0.4371 - key_acc176/197 [=========================>....] - ETA: 11s - loss: 2.9001 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6092 - quality_loss: 0.5295 - inversion_loss: 0.5369 - root_loss: 0.4371 - key_acc177/197 [=========================>....] - ETA: 11s - loss: 2.9002 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6092 - quality_loss: 0.5295 - inversion_loss: 0.5369 - root_loss: 0.4371 - key_acc178/197 [==========================>...] - ETA: 10s - loss: 2.9003 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6093 - quality_loss: 0.5295 - inversion_loss: 0.5369 - root_loss: 0.4371 - key_acc179/197 [==========================>...] - ETA: 10s - loss: 2.9004 - key_loss: 0.4640 - degree_1_loss: 0.3234 - degree_2_loss: 0.6093 - quality_loss: 0.5295 - inversion_loss: 0.5370 - root_loss: 0.4372 - key_acc180/197 [==========================>...] - ETA: 9s - loss: 2.9004 - key_loss: 0.4640 - degree_1_loss: 0.3235 - degree_2_loss: 0.6094 - quality_loss: 0.5296 - inversion_loss: 0.5370 - root_loss: 0.4372 - key_accu181/197 [==========================>...] - ETA: 9s - loss: 2.9005 - key_loss: 0.4640 - degree_1_loss: 0.3235 - degree_2_loss: 0.6094 - quality_loss: 0.5296 - inversion_loss: 0.5370 - root_loss: 0.4372 - key_accu182/197 [==========================>...] - ETA: 8s - loss: 2.9006 - key_loss: 0.4640 - degree_1_loss: 0.3235 - degree_2_loss: 0.6094 - quality_loss: 0.5296 - inversion_loss: 0.5370 - root_loss: 0.4372 - key_accu183/197 [==========================>...] - ETA: 7s - loss: 2.9007 - key_loss: 0.4640 - degree_1_loss: 0.3235 - degree_2_loss: 0.6095 - quality_loss: 0.5296 - inversion_loss: 0.5371 - root_loss: 0.4372 - key_accu184/197 [===========================>..] - ETA: 7s - loss: 2.9008 - key_loss: 0.4640 - degree_1_loss: 0.3235 - degree_2_loss: 0.6095 - quality_loss: 0.5296 - inversion_loss: 0.5371 - root_loss: 0.4372 - key_accu185/197 [===========================>..] - ETA: 6s - loss: 2.9009 - key_loss: 0.4640 - degree_1_loss: 0.3235 - degree_2_loss: 0.6095 - quality_loss: 0.5297 - inversion_loss: 0.5371 - root_loss: 0.4372 - key_accu186/197 [===========================>..] - ETA: 6s - loss: 2.9010 - key_loss: 0.4639 - degree_1_loss: 0.3235 - degree_2_loss: 0.6096 - quality_loss: 0.5297 - inversion_loss: 0.5371 - root_loss: 0.4372 - key_accu187/197 [===========================>..] - ETA: 5s - loss: 2.9011 - key_loss: 0.4639 - degree_1_loss: 0.3235 - degree_2_loss: 0.6096 - quality_loss: 0.5297 - inversion_loss: 0.5372 - root_loss: 0.4372 - key_accu188/197 [===========================>..] - ETA: 5s - loss: 2.9012 - key_loss: 0.4639 - degree_1_loss: 0.3235 - degree_2_loss: 0.6097 - quality_loss: 0.5297 - inversion_loss: 0.5372 - root_loss: 0.4372 - key_accu189/197 [===========================>..] - ETA: 4s - loss: 2.9013 - key_loss: 0.4639 - degree_1_loss: 0.3235 - degree_2_loss: 0.6097 - quality_loss: 0.5297 - inversion_loss: 0.5372 - root_loss: 0.4372 - key_accu190/197 [===========================>..] - ETA: 3s - loss: 2.9014 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6097 - quality_loss: 0.5298 - inversion_loss: 0.5372 - root_loss: 0.4372 - key_accu191/197 [============================>.] - ETA: 3s - loss: 2.9014 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6098 - quality_loss: 0.5298 - inversion_loss: 0.5372 - root_loss: 0.4372 - key_accu192/197 [============================>.] - ETA: 2s - loss: 2.9015 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6098 - quality_loss: 0.5298 - inversion_loss: 0.5373 - root_loss: 0.4372 - key_accu193/197 [============================>.] - ETA: 2s - loss: 2.9016 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6099 - quality_loss: 0.5298 - inversion_loss: 0.5373 - root_loss: 0.4372 - key_accu194/197 [============================>.] - ETA: 1s - loss: 2.9017 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6099 - quality_loss: 0.5298 - inversion_loss: 0.5373 - root_loss: 0.4372 - key_accu195/197 [============================>.] - ETA: 1s - loss: 2.9018 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6099 - quality_loss: 0.5299 - inversion_loss: 0.5373 - root_loss: 0.4372 - key_accu196/197 [============================>.] - ETA: 0s - loss: 2.9019 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6100 - quality_loss: 0.5299 - inversion_loss: 0.5373 - root_loss: 0.4372 - key_accu197/197 [==============================] - ETA: 0s - loss: 2.9020 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6100 - quality_loss: 0.5299 - inversion_loss: 0.5374 - root_loss: 0.4372 - key_accu197/197 [==============================] - 113s 573ms/step - loss: 2.9021 - key_loss: 0.4639 - degree_1_loss: 0.3236 - degree_2_loss: 0.6100 - quality_loss: 0.5299 - inversion_loss: 0.5374 - root_loss: 0.4372 - key_accuracy: 0.8537 - degree_1_accuracy: 0.9109 - degree_2_accuracy: 0.7909 - quality_accuracy: 0.8203 - inversion_accuracy: 0.7927 - root_accuracy: 0.8662 - val_loss: 4.3752 - val_key_loss: 0.7263 - val_degree_1_loss: 0.3661 - val_degree_2_loss: 0.8904 - val_quality_loss: 0.8303 - val_inversion_loss: 0.8639 - val_root_loss: 0.6982 - val_key_accuracy: 0.7659 - val_degree_1_accuracy: 0.9091 - val_degree_2_accuracy: 0.7012 - val_quality_accuracy: 0.7368 - val_inversion_accuracy: 0.6695 - val_root_accuracy: 0.7888

loss: 2.9021
key_loss: 0.4639
degree_1_loss: 0.3236
degree_2_loss: 0.6100
quality_loss: 0.5299
inversion_loss: 0.5374
root_loss: 0.4372

key_accuracy: 0.8537
degree_1_accuracy: 0.9109
degree_2_accuracy: 0.7909
quality_accuracy: 0.8203
inversion_accuracy: 0.7927
root_accuracy: 0.8662

val_loss: 4.3752
val_key_loss: 0.7263
val_degree_1_loss: 0.3661
val_degree_2_loss: 0.8904
val_quality_loss: 0.8303
val_inversion_loss: 0.8639
val_root_loss: 0.6982

val_key_accuracy: 0.7659
val_degree_1_accuracy: 0.9091
val_degree_2_accuracy: 0.7012
val_quality_accuracy: 0.7368
val_inversion_accuracy: 0.6695
val_root_accuracy: 0.7888

The output of analyse_results:

{'key': 76.59036378963751, 
'degree 1': 90.9149860579729, 
'degree 2': 70.11866934699435, 
'quality': 73.67874975682511, 
'inversion': 66.94766876337462, 
'root': 78.87945010051229, 
'degree': 64.34083392776084, 
'secondary': 5.949008498583569, 
'derived root': 68.15381622462876, 
'roman': 50.10051228843785, 
'roman + inv': 34.7383438168731, 
'root coherence': 73.43881719732832, 
'd7 no inv': 22.06266318537859}

Re-running the last experiment with a higher patience on EarlyStopping

loss: 2.2859
key_loss: 0.3149
degree_1_loss: 0.2513
degree_2_loss: 0.4774
quality_loss: 0.4372
inversion_loss: 0.4424
root_loss: 0.3627

key_accuracy: 0.8972
degree_1_accuracy: 0.9269
degree_2_accuracy: 0.8354
quality_accuracy: 0.8508
inversion_accuracy: 0.8309
root_accuracy: 0.8864

val_loss: 4.3278
val_key_loss: 0.7584
val_degree_1_loss: 0.3218
val_degree_2_loss: 0.9097
val_quality_loss: 0.8227
val_inversion_loss: 0.8614
val_root_loss: 0.6537

val_key_accuracy: 0.7573
val_degree_1_accuracy: 0.9145
val_degree_2_accuracy: 0.7081
val_quality_accuracy: 0.7501
val_inversion_accuracy: 0.6836
val_root_accuracy: 0.8085

The output of analyse_results:

{'key': 75.73438817197328, 
'degree 1': 91.45321315089812, 
'degree 2': 70.812528370404, 
'quality': 75.00810582971273, 
'inversion': 68.36132546527463, 
'root': 80.85078788664808, 
'degree': 64.9503923221581, 
'secondary': 11.189801699716714, 
'derived root': 70.20945463977692, 
'roman': 50.83976395823876, 
'roman + inv': 35.620258089618055, 
'root coherence': 77.66033331171779, 
'd7 no inv': 21.148825065274153}

Mark and Ich are both optimistic about the outcomes so far. Great news!

Things to try next:

[ ] Event-based representation instead of fixed timestep
[ ] Texturization of augmented examples

The event-based representation is something I'll try on my network first, because it's going to be much easier that way.

napulen / phd_thesis