mrc-cbt

A Tensorflow implementation of ASReader: "Text Comprehension with the Attention Sum Reader Network" (ACL2016) https://arxiv.org/abs/1603.01547

The code adopted from https://github.com/NLPLearn/QANet

tensorflow 1.12 python 3.6 tqdm nltk

step 0:

download CBT dataset and pre-trained glove file, and put the to the right dir:

datasets/cbt/
			cbtest_NE_train.txt
			cbtest_NE_valid_2000ex.txt
			cbtest_NE_test_2500ex.txt
			cbtest_CN_train.txt
			cbtest_CN_valid_2000ex.txt
			cbtest_CN_test_2500ex.txt
datasets/glove/
			glove.840B.300d.txt

Step 1:

convert the .txt file to json:
python cbt_text2json.py

Step 2:

prepare the TRRecord file for train/valid/test
python config.py --mode prepro

Step 3:

python config.py --mode train

Step 4:

python config.py --mode test

Run tensorboard for visualisation.

$ tensorboard --logdir=./ --port=9102