lizuyao2010/lstm_theano

summer at ttic project

#run code python lstm_adagrad.py

lstm theano code for question answering with freebase

##encode question with lstm encoder hidden state size 150, embedding size 150
question embedding is sum over hiddent states of lstm encoder
answer embedding is sum over path(relation) embeddings
##loss funciton margin ranking loss which seprate correct answer score with incorrect answer score with at least margin 1. ##optimizier adagrad which is a parameter learning rate adaption method.