MXNET-SMN

Sequential Matching Network(ACL 2017) reimplemented by MXNET1.0-GPU//but you can always use cpu

mxnetSMNACL 2017MXNET1.0

What's SMN

SMN

SoTA model of multi-turn retrieval-based conversation systems.

You can see the paperSMN

Don't forget reference it if you use.

Wu, Yu, et al. "Sequential Matching Network: A New Archtechture for Multi-turn Response Selection in Retrieval-based Chatbots." ACL. 2017.

SMN.

SMN.

SMNbaselineACL

How to use?

1.get data from msradouban corpus

2.get pre-trained word2vec.model use bash python3.5 gen_w2v.py train.txt train_vec.model train_vec.vec

3.get processed data use Process.py(also only support py3)

4.run the model

1. douban corpus

2.Embeddingpython3.5 gen_w2v.py train.txt train_vec.model train_vec.vec

3.Process.py

4.

py

| | | |:------------------------------------:|------------------------------------| |gen_w2v.py|| |Process.py|| |model.py||

Params of Model

batch_size = 1000(with 1 titan xp)

embedding_size = 200

gru_layer = 1

max_turn = 10

max_len = 50

lr = 0.001

Why you use MXNET

MXNET

Fast!

**!

MXNET -

Other versions?

TensorFlow,

There is Theano version.

theanomxnet

**

Your code is not pythonic!

shit

Sorry to hear that.

pythonic

Can I use your code to do a chatbot

If you can use lucene then you can.

querylucene

How to tipping

You can sent your money to this website

E

TODO:

1.save model

2.add dropout

3.add evaluations

4.add inference