CNN QA(Question and Answer)问答的theano和tensorflow代码

3420阅读 0评论2016-06-03 jiangwen127
分类:大数据





================result==================

theano and tensorflow cnn code for insuranceQA

theano code, test1 top-1 precision : 61.5% (see ./insuranceQA/acc) tensorflow code, test1 top-1 precision : 62.6%

the best precision in the paper is 62.8% (see Applying Deep Leaarning To Answer Selection: A study and an open task)

================dataset================

dataset is large, only test1 sample is given (see ./insuranceQA/test1.sample)

I converted original idx_xx format to real-word format (see ./insuranceQA/train ./insuranceQA/test1.sample)

you can get the original dataset from 

word embedding is trained by word2vec toolkit

=================run=====================

reformat the original dataset(see my train and test1.sample)
change filepath to your dataset(see TODO in insqa_cnn.py)
python insqa_cnn.py


上一篇:机器学习实践中应避免的七种常见错误
下一篇:An overview of gradient descent optimization algorithms