Remaining issues with additional datasets

[yoavkatz]

There are few open issues:

There is no multi_label template (fix required to unfair_tos and reuters) Can I use text_type : argument? I wonder if dbpedia_14 is of type text or paragraph I do not remember why I need to add ListFieldValues(fields=["label"], to_field="label") , maybe it's not relevant anymore. There are some datasets where we played with splits. E.g. law_stack_exchange.py in we swapped train and test. Is it ok to do it? Or in financial_tweets validation went to test.

[elronbandel]

Also I think its worth changing the backbone template of classification tasks from InputOutputTemplate to MultipleChoiceTemplate so we can allow models to choose from the options field it generates

[yoavkatz]

I think MultipleChoiceTemplate is an additional possible template. I think it's only relevant for multi class and not mult label.