mistakes with the "retrain/train.py" code

[TomatoChocolate12]

The labels that are used for training the model do not correspond to the correct values. The last column of the featurized tensor is being used instead and this corresponds to interatomic interactions (features that are extracted basically). The labels dataframe labels_df which is also already loaded, should be used instead.

[zchwang]

You can input the features and labels as two separate files, which might be a simpler approach.