temp fix for when forward returns fewer probs than input; related to issue 9

adding training documentation

adding training documentation

Update TODO.md

minor typo in column name for eval mode

changing requirement to be just sentence and sentid

clarifying ROI is zero-indexed

fixed bug in by_cond aggregation

fixing error with punctuation handling; adding word column for by_word

linking more detailed documentation

adding documentation for experiments

adding token classification

updating todo

basic token classification implementation

adding token classification args