FusionNet

This is my pytorch implementation of the FusionNet: Fusing via Fully-aware Attention with Application to Machine Comprehension for chinese machine comprehension dataset (not public dataset). However, I am now implementing the preprocessing/decoding for SQuAD, I think the SQuAD compatiable version will be released soon.

Feel free to use/modify it. Any improvement will be appreciated.

HOW TO RUN:

$ python3 preprocess.py
$ python3 train.py

TODO:

I will implement following word feature soon

• Glove word vector and fine-tune (Maybe fasttext?)
• CoVe vector
• POS, NER, normalized term frequency
• context word appears in the quesiton?

Result:

Kaggle competition

References:

• FUSION NET: FUSING VIA FULLY-AWARE ATTENTION WITH APPLICATION TO MACHINE COMPREHENSION
• Awesome implementation of the DrQA