akariasai / document-qa-mlqa Goto Github PK

Need to add a pred mode which takes paragraph and question pairs and return the answers from model.
Currently, like the baseline model, I just add our test cases to test corpus options, but it should be great we don't need to run preprocessing over and over to experiment just 50 questions.

Things to do :

• Add args option for test corpus, and arbitrary corpus file path for docqa/eval/squad_eval.py
• Add corpus.get_pred(args.pred_filepath) in order to convert the input files to the appropriate question/paragraph dataset the document QA model. the copus is SquadCorpus, which is defined in docqa/squad/squad_data.py
• The pickle file need to be created for the loading, so add function to craete the same format dataset by adding create_pred_corpus to docqa/squad/build_squad_dataset.py