Layer count mismatch - Githubissues

lewis-chinery / Humatch

Fast, gene-specific joint humanisation of antibody heavy and light chains.

BSD 3-Clause "New" or "Revised" License

1 stars 4 forks source link

Layer count mismatch #1

Open QUEST2179 opened 1 month ago

QUEST2179 commented 1 month ago

tried the example in notebook folder

(20, 3) is_human heavy light 0 1 QVQ-LVQSGA-EVNKPGASVKVSCKASGYTF---------TGYVVH... EIV-LTQSPVTLSLSPGERATLSCRASQSV-----------SFYLA... 1 1 QVQ-LVQSGA-EVKKPGASVKVSCKASGYTF---------TGYYMH... SYE-LTQPPS-VSVSPGQTARITCSGDALP-----------KQYAY... Traceback (most recent call last): File "C:\Users\work\antibody\Humatch-master\testClassify.py", line 25, in cnn_heavy = load_cnn(os.path.join(weights_dir, "heavy.weights.h5"), "heavy") File "C:\Users\work\antibody\Humatch-master\Humatch\model.py", line 130, in load_cnn CNN.load_weights(weights) File "C:\Users\miniconda3\envs\gpu9\lib\site-packages\keras\utils\traceback_utils.py", line 70, in error_handler raise e.with_traceback(filtered_tb) from None File "C:\Users\miniconda3\envs\gpu9\lib\site-packages\keras\saving\hdf5_format.py", line 817, in load_weights_from_hdf5_group raise ValueError( ValueError: Layer count mismatch when loading weights from file. Model expected 3 layers, found 0 saved layers.

lewis-chinery commented 1 month ago

Hi, I just tried deleting everything, reinstalling, and running the notebooks and it seemed to work OK for me (the weights download automatically). Could you please provide a bit more information:

Which notebook did you try to run?
Which cell did the notebook fail at?
How did you set up your environment? (It looks like you're perhaps using conda and not venv)
How are you running your notebook? (I am using VSCode and running on Linux Fedora 38)
Did you create the testClassify.py file and what does this look like? (It seems like this is being called too)

If you do need to use a different set-up to that described in the README (especially using conda and not venv) then you could try manually downloading the weights as described in the README and trying again

jesperdlau commented 3 weeks ago

I had the same error and fixed it by upgrading Tensorflow to 2.17. Then I got a different error, something like: TypeError: generator yielded an element that did not match the expected structure. The expected structure was (tf.float64,), but the yielded element was [[[-0.47 0.24 0.07 ...

This I fixed by changing the way the data generator is used inside the classify script, for the predict_from_list_of_seq_strs function. Instead of return model.predict(test_generator, verbose=CNN_verbose) I used: return model.predict(next(iter(test_generator)), batch_size=batch_size, verbose=CNN_verbose) (Remember to "pip install ." after changing code) I other words, I parse the data itself, not the generator. I seems like the custom generator isn't compatible with Tensorflow 2.17 at least. Hope it makes sense. So far it works, but might not be the case always. It took me a long time to figure out.

@lewis-chinery Can you please share the exact package versions you have?