bloomsburyai/question-generation

What is the model trained on ?

Closed this issue · 4 comments

Is it with Squad1.1 or 2.0?

The questions added in squad 2 are adversarial and don't have answers, so can't be used to train the model

Which data is more appropriate for training is it squad 1.1 or 2.0 as I see your code having the flexibility to do this and I tweaked these lines and i was able to start training

if qa['is_impossible']:
el = (para['context'], qa['question'], qa['plausible_answers'][0]['text'] if not dev else "", int(qa['plausible_answers'][0]['answer_start']) if not dev else None, True)
else:
el = (para['context'], qa['question'], qa['answers'][0]['text'], int(qa['answers'][0]['answer_start']), False)

But not sure whether that would help or get worse? please suggest .

Changes:

  1. Basically neglecting the last parameter(5th one)
  2. adding if not qa['is_impossible']: and commenting the else.

The questions added in squad 2 are adversarial and don't have answers, so can't be used to train the model

Got it thanks.