Make train-test split accessible

[MaksHess]

While the current method of creating a train-test split based on the hash of the index-column of the underlying DataFrame always leads to the same splits, even when adding new data, the underlying splits are not easily accessible to the user. The easiest solution would be to add an additional column 'is_train' to the classifier, that is updated every time Classifier.train_test_split is called.