Bug (I think) in "peptide_coefficient_predictor.py" - proposed partial solution

[Arthfael]

Hi, and sorry for the slow trickle of reports, I am testing this in between projects and as time allows.

I am pretty new to python, but still I am pretty sure that I have found a bug in "peptide_coefficient_predictor.py": iterating the script over the datasets provided in the package, using the parameters from train_models.sh (same directory), it failed for some but not all datasets. After investigating, I realized that when splitting the data between train_runs and test_runs for all datasets with odd numbers of samples, it was trying to access a non existing column number. The problem did not occur when there was an even number of samples.

Proposed partial fix:
After line

test_runs = np.array([[2*i, 2*i+1] for i in test_runs]).astype(int).ravel()

add:

train_runs = train_runs[train_runs+1 <= q_df.shape[1]]
test_runs = test_runs[test_runs+1 <= q_df.shape[1]]

I have tested it. I say it is a partial fix in that it fixes the immediate error, however those datasets still fail: first, I get the following warning:

>>> model = define_model()
... (some messages)
WARNING:tensorflow:Output {0} missing from loss dictionary. We assume this was done on purpose. The fit and evaluate APIs will not be expecting any data to be passed to custom_loss_layer_7.

then the following error:

>>> patience_count = 0
>>> 
>>> for epoch in range(1000):
...     
...     history = model.fit(x = [X_train, C_train,  joined_intensities], 
...                         y = None, 
...                         epochs=1, batch_size = batch_size) 
... (etc)
AttributeError: 'Adam' object has no attribute 'get_updates'
TypeError: object of type 'NoneType' has no len()

I do not know yet whether they are related. Anyway, I will investigate further in the next few days.

[Arthfael]

Actually, the solution seemed only partial because I was using reticulate and running the script through RStudio. From Window's command line, it works for all datasets tested, so seems to be complete.