Machine Reading Comprehension on DuReader

Using BiDAF and QANet on DuReader. Writen by YanXu, FangYueran and ZhangTianyang

Pretrained embedding

When we train the QANet model, we use the pretrained word embedding from Baidu Encyclopedia, you can down load and save in folder ./embedding

Full experimental results

Complete experimental results (including data sets, log of experimental records, tensorboard, and predicted output) can be downloaded from the Baidu network disk：https://pan.baidu.com/s/1qoxnF00wyJ2dqcAPDYTb8w code：gn5b, You can override it with the ./data

You can also train and test in completed DuReader dataset http://ai.baidu.com/broad/download

Usage

BiDAF

Generate dict and embedding：python BaiduRun.py --prepare
Train： python BaiduRun.py --train
Evaluate on dev： python BaiduRun.py --evaluate
Output the answers： python BaiduRun.py --predict

QANet

Generate dict and embedding：python OurRun.py --prepare
Train： python OurRun.py --train
Evaluate on dev： python OurRun.py --evaluate
Output the answers： python OurRun.py --predict