Error when running Evaluate : axis 1 is out of bounds for array of dimension 0

[Julmatap]

Hello everyone,

Firstly I would like to say thank you for this amazing project and the release of your code. I am very new to programming and ML and I wanted to try your project.

I downloaded the data, installed the environments on latest versions, followed the reproduced results and looked at the issues and tried to solve the errors I got by myself as much as possible, but right now i'm stucked.

I first prepared masks & metadata then I downloaded your best models weights, created the transformers folder in data_experiments/mapping_challenge_baseline and copied "unet" and "scoring_model" there. I then changed the values in neptume.yaml as suggested and tried to evaluate by running : "python main.py evaluate --pipeline_name unet"

Here is what I got :

Hope I gave you enough informations, and that you can help me resolve this issue.

Again, thank you guys !

PS : I saw that issue #228 has this problem also, but i don't have the "valid data is None" as you can see on screenshots here attached

[jakubczakon]

Hi @Julmatap, thank for the nice words.

Could you please paste your directory structure (data paths) and the content of the neptune.yaml file (data paths) ?

[Julmatap]

Sure ! (Thanks for your quick reply !)

Based on yours, here's mine :

Data_OSM_Buildings:
|--   README.md
|-- ...
|-- data_raw
    |-- train 
         |-- images 
         |-- annotation.json
    |-- val 
         |-- images 
         |-- annotation.json
    |-- test (and not test_images) 
         |-- img1.jpg
         |-- img2.jpg
         |-- ...
|-- data_meta
    |-- masks_overlayed_eroded_{}_dilated_{} 
         |-- train 
             |-- distances 
             |-- masks 
             |-- sizes 
         |-- val 
             |-- distances 
             |-- masks 
             |-- sizes 
    |-- metadata.csv
|-- data_experiments
    |-- mapping_challenge_baseline 
         |-- transformers
             |--unet
             |--scoring_model
         |-- outputs 
         |-- tmp

And the parameters content of my neptune.yaml looks like this : project: shared/showroom

name: mapping_challenge_baseline tags: [solution_5]

parameters:

Data Paths

data_dir: data_raw meta_dir: data_meta masks_overlayed_prefix: masks_overlayed experiment_dir: data_experiments/mapping_challenge_baseline

Hope I understood what you asked me.

[Julmatap]

I'll try to use the exact same directory structure as you did in your example and correct the neptune.yaml accordingly to see if it resolves the problem. I didn't find where in the code it would pose a problem to have this kind of structure but since i'm a beginner it may be normal lmao.

[jakubczakon]

Mhm, so do you have data_raw data_meta and others inside of the data directory or a the same level as README.md ?

[Julmatap]

Dear Jakub, I had data_raw and data_meta folders at the same level as README.md . After you answered I made the directory structure exactly like what you suggested which now looks like this :

Data_OSM
|--   README.md
|-- ...
|-- data
    |-- raw
         |-- train 
            |-- images 
            |-- annotation.json
         |-- val 
            |-- images 
            |-- annotation.json
         |-- test_images 
            |-- img1.jpg
            |-- img2.jpg
            |-- ...
    |-- meta
         |-- masks_overlayed_eroded_0_dilated_0 
            |-- train 
                |-- distances 
                |-- masks 
                |-- sizes 
            |-- val 
                |-- distances 
                |-- masks 
                |-- sizes 
    |-- experiments
        |-- mapping_challenge_baseline
            |-- checkpoints 
            |-- transformers 
                |--unet
                |--scoring_model
            |-- outputs 

I changed my neptume.yaml to look like this :

Data Paths

data_dir: data/raw meta_dir: data/meta masks_overlayed_prefix: masks_overlayed experiment_dir: data/experiments/mapping_challenge_baseline

However, when i run evaluate, it still throws me the exact same error.

[jakubczakon]

Could you try and run the training pipeline first?

python main.py train --pipeline_name unet

[Julmatap]

I did it as soon as you told me to, but it doesn't seem like it's running, here's a screenshot :

[Julmatap]

Could it be linked to my environment ? I had to change versions and some stuff for it to work (i'm on windows10 x64). I couldn't install Torch v 0.3.1 and others via your environment.yml.

Here is my complete env for information :

$ conda list
# packages in environment at C:\Users\matierej\Anaconda3\envs\mapping:
#
# Name                    Version                   Build  Channel
attrdict                  2.0.1                    pypi_0    pypi
attrs                     19.3.0                     py_0
backcall                  0.2.0                      py_0
bleach                    3.1.5                      py_0
bravado                   10.6.2                   pypi_0    pypi
bravado-core              5.17.0                   pypi_0    pypi
ca-certificates           2020.6.24                     0
certifi                   2020.6.20                py38_0
chardet                   3.0.4                    pypi_0    pypi
click                     7.1.2                    pypi_0    pypi
colorama                  0.4.3                      py_0
cycler                    0.10.0                   pypi_0    pypi
cython                    0.29.21                  pypi_0    pypi
decorator                 4.4.2                      py_0
defusedxml                0.6.0                      py_0
entrypoints               0.3                      py38_0
future                    0.18.2                   pypi_0    pypi
gitdb                     4.0.5                    pypi_0    pypi
gitpython                 3.1.7                    pypi_0    pypi
idna                      2.10                     pypi_0    pypi
imageio                   2.9.0                    pypi_0    pypi
imgaug                    0.4.0                    pypi_0    pypi
importlib-metadata        1.7.0                    py38_0
importlib_metadata        1.7.0                         0
ipykernel                 5.3.4            py38h5ca1d4c_0
ipython                   7.17.0                   pypi_0    pypi
ipython-genutils          0.2.0                    pypi_0    pypi
ipython_genutils          0.2.0                    py38_0
jedi                      0.17.2                   pypi_0    pypi
jinja2                    2.11.2                     py_0
joblib                    0.16.0                   pypi_0    pypi
jsonpointer               2.0                      pypi_0    pypi
jsonref                   0.2                      pypi_0    pypi
jsonschema                3.2.0                    py38_0
jupyter_client            6.1.6                      py_0
jupyter_core              4.6.3                    py38_0
kiwisolver                1.2.0                    pypi_0    pypi
libsodium                 1.0.18               h62dcd97_0
lightgbm                  2.3.1                    pypi_0    pypi
m2w64-gcc-libgfortran     5.3.0                         6
m2w64-gcc-libs            5.3.0                         7
m2w64-gcc-libs-core       5.3.0                         7
m2w64-gmp                 6.1.0                         2
m2w64-libwinpthread-git   5.0.0.4634.697f757               2
markupsafe                1.1.1            py38he774522_0
matplotlib                3.3.0                    pypi_0    pypi
mistune                   0.8.4           py38he774522_1000
monotonic                 1.5                      pypi_0    pypi
msgpack                   1.0.0                    pypi_0    pypi
msgpack-python            0.5.6                    pypi_0    pypi
msys2-conda-epoch         20160418                      1
munch                     2.5.0                    pypi_0    pypi
nbconvert                 5.6.1                    py38_0
nbformat                  5.0.7                      py_0
neptune-client            0.4.119                  pypi_0    pypi
neptune-contrib           0+unknown                pypi_0    pypi
networkx                  2.4                      pypi_0    pypi
notebook                  6.0.3                    py38_0
numpy                     1.19.1                   pypi_0    pypi
oauthlib                  3.1.0                    pypi_0    pypi
opencv-python             4.3.0.36                 pypi_0    pypi
openssl                   1.1.1g               he774522_0
packaging                 20.4                       py_0
pandas                    1.1.0                    pypi_0    pypi
pandoc                    2.10                          0
pandocfilters             1.4.2                    py38_1
parso                     0.7.1                    pypi_0    pypi
pickleshare               0.7.5                 py38_1000
pillow                    7.2.0                    pypi_0    pypi
pip                       20.1.1                   py38_1
pretrainedmodels          0.7.4                    pypi_0    pypi
prometheus_client         0.8.0                      py_0
prompt-toolkit            3.0.5                      py_0
psutil                    5.7.2                    pypi_0    pypi
py3nvml                   0.2.6                    pypi_0    pypi
pycocotools-windows       2.0.0.2                  pypi_0    pypi
pydensecrf                1.0rc2                   pypi_0    pypi
pydot-ng                  2.0.0                    pypi_0    pypi
pygments                  2.6.1                      py_0
pyjwt                     1.7.1                    pypi_0    pypi
pyparsing                 2.4.7                      py_0
pyrsistent                0.16.0           py38he774522_0
python                    3.8.5                he1778fa_0
python-dateutil           2.8.1                      py_0
pytz                      2020.1                   pypi_0    pypi
pywavelets                1.1.1                    pypi_0    pypi
pywin32                   227              py38he774522_1
pywinpty                  0.5.7                    py38_0
pyyaml                    5.3.1                    pypi_0    pypi
pyzmq                     19.0.1           py38ha925a31_1
requests                  2.24.0                   pypi_0    pypi
requests-oauthlib         1.3.0                    pypi_0    pypi
rfc3987                   1.3.8                    pypi_0    pypi
scikit-image              0.17.2                   pypi_0    pypi
scikit-learn              0.23.2                   pypi_0    pypi
scipy                     1.5.2                    pypi_0    pypi
send2trash                1.5.0                    py38_0
setuptools                49.2.0                   py38_0
shapely                   1.7.0                    pypi_0    pypi
simplejson                3.17.2                   pypi_0    pypi
six                       1.15.0                     py_0
smmap                     3.0.4                    pypi_0    pypi
sqlite                    3.32.3               h2a8f88b_0
strict-rfc3339            0.7                      pypi_0    pypi
swagger-spec-validator    2.7.3                    pypi_0    pypi
terminado                 0.8.3                    py38_0
testpath                  0.4.4                      py_0
threadpoolctl             2.1.0                    pypi_0    pypi
tifffile                  2020.7.24                pypi_0    pypi
torch                     1.4.0+cpu                pypi_0    pypi
torchvision               0.5.0+cpu                pypi_0    pypi
tornado                   6.0.4            py38he774522_1
tqdm                      4.48.2                   pypi_0    pypi
traitlets                 4.3.3                    py38_0
typing-extensions         3.7.4.2                  pypi_0    pypi
urllib3                   1.25.10                  pypi_0    pypi
vc                        14.1                 h0510ff6_4
vs2015_runtime            14.16.27012          hf0eaf9b_3
wcwidth                   0.2.5                      py_0
webcolors                 1.11.1                   pypi_0    pypi
webencodings              0.5.1                    py38_1
websocket-client          0.57.0                   pypi_0    pypi
wheel                     0.34.2                   py38_0
wincertstore              0.2                      py38_0
winpty                    0.4.3                         4
xgboost                   1.1.1                    pypi_0    pypi
xmltodict                 0.12.0                   pypi_0    pypi
zeromq                    4.3.2                ha925a31_2
zipp                      3.1.0                      py_0
zlib                      1.2.11               h62dcd97_4

[jakubczakon]

Ok, I see.

I think that is the problem. Everything was written for torch==0.3.1 and newer releases (torch==1.4 for sure) have changed things. Could you try and install it by hand:

pip install torch==0.3.1

[Julmatap]

I tried again installing torch==0.3.1 and here's what I got

[jakubczakon]

I think you need to downgrade python for this. As environment.yml suggest it should be python=3.6.8 The easiest way to do it is to go:

conda env create -f environment.yml

but as I understand that isn't working. In that case I'd just create a clean conda environment with python 3.6:

conda create -n py_36_env python=3.6

activate it

conda activate py_36_env

and then install the dependencies from environment.yml

[Julmatap]

Dear @jakubczakon thank you for your time helping me !

It was indeed a problem with Torch version.

However I still couldn't install Torch==0.3.1 even after downgrading python and so on... I tried manually installing it, but I still had the same error as per my previous screenshot, I also tried finding a .whl but there was none for v 0.3.1 win64. So I tried using the closest version I could find and ended up using Torch version 0.4.1, and it seems to be working just fine.

Right now I'm still evaluating since more than 5 hours, and I can tell it's running with neptune and my memory usage. (But it's not using GPU despite my GPU being cuda compatible, is it normal ?)

Once it is finished and I am sure it is working, I will post my conda list, if it can help someone else facing this issue.

Again, thank you very much for your time 👍

[jakubczakon]

That is awesome thank you @Julmatap!

You can see if it is running but going to the Charts or Logs section of the UI. You should see some activity there.

[jakubczakon]

Also, it seems that there is a message in the terminal that explains why the GPU is not logged -> sending GPU metrics was blocked by the system. This is very much unexpected.

[Julmatap]

Unfortunately there is nothing "No charts here" and "waiting for data", and I think there's a problem, someone else launched an experiments 3mn and he has already some stuffs.

[jakubczakon]

You are running the evaluation now correct? If so can you try running training and see if something is happening?

Also you can go to your terminal and run:

nvidia-smi

to see if your GPU is actually doing something.

[Julmatap]

Hi Jakub,

Yes I was running the evaluation, I tried running the training today, and there's still no charts. I runned nvidia-smi before and today again and it says that no processes is running.

I guess I'll just start again on a Linux VM. Has anyone already suceeded in running it on windows ? The operating system blocking the request is something I don't find any informations on.

[jakubczakon]

I see @Julmatap,

Unfortunately, I don't know if anyone succeeded on Windows -> everyone I know who used this repo was using Linux.

[Julmatap]

I confirm it is working on Linux, I just did an evaluation and it worked, just had to wait a long time at "steps >>> step unet transforming...".

Again, thanks for your time @jakubczakon ! 👍

[jakubczakon]

That is awesome!

I only wish I could be of more help but I am proud of you getting it done.

[Julmatap]

Dear Jakub,

One last question :

I have my metrics which looks like this :

However, when I run the notebook "results on exploration" my predictions are blank :

Edit : I did open predictions.json and manually added the first image_id I saw on the notebook and the prediction runs fine. However it seems that all the val folder has not been predicted and so when I run random choices it usually doesn't work.