pablo-unzueta/dml-nmr

dml-nmr, python scripts for nuclear magnetic resonance machine learning

dml-nmr's Issues

bug in ensemble.py

The refactored code does not load the models correctly. It tries to load model_0_n1_8.h5 instead of model_10_n1_8.h5

I suggest the following corrected code:

def load_weights(self, atom_type):
saved_models_dir = os.path.join(self.path, atom_type)

     model_paths = [
        os.path.join(saved_models_dir, f"model_{i+1}_n1_8.h5") for i in range(10)
    ]
  ....

incompatibility with latest tensorflow/numpy libaries

After upgrading the system to newer libraries (2.11 for tensorflow, 1.24 for numpy), the following line(s) in ensemble.py throw an error:

ensemble_ml_corrs_H = np.asarray(ensemble_ml_corrs_H)

I could not quite get to the bottom why it worked before but it isn't working now because the output from 'predict' is not a list with uniform dimensions.

The problem goes away and previous results are generated when the line (and corresponding lines for other atoms) are commented out but better code could be more robust to work with different versions.