Unwanted behaviour when batch doesn't have all categories

[Aure20]

Hi, I just wanted to point out a small corner case I encountered while using your code, in basemodel.py, when you are appending the scores to the train_result, what might happen is that the current batch is missing a class (i.e. if it's binary classification y_true has only 0 in it for example), this creates a problem when using the cross_entropy and auc_score as they need all classes to be present in each batch so I corrected the corss_entropy case by specifying the classes with the labels parameter, but the same doesn't seem to work with the auc case for which I implemented a try and catch.