jsrimr / tensorflow-memn2n

End-To-End Memory Networks

TensorFlow-MemN2N

End-To-End Memory Networks

TensorFlow version == 1.4

paper

* #https://arxiv.org/abs/1503.08895

dataset

* babi (tasks_1-20_v1-2) 
    * https://research.fb.com/downloads/babi/

MemN2N.py

* End to End Memory Network를 구현한 class

babi_process.py

* babi dataset을 전처리 하는 코드.

training.py

* 전처리된 babi dataset을 학습하는 코드.

result

Jointly Learning
Model A: Embedding size = 50, Adjacent, Position Encoding, Temporal Encoding
Model B: Embedding size = 100, Adjacent, Position Encoding, Temporal Encoding, Relu

Task	Model A	Model B
1	0.998	0.999
2	0.828	0.756
3	0.7	0.788
4	0.991	0.998
5	0.935	0.983
6	0.983	1.0
7	0.888	0.951
8	0.917	0.966
9	0.989	1.0
10	0.977	0.999
11	0.999	1.0
12	0.973	0.999
13	0.999	1.0
14	1.0	0.996
15	1.0	1.0
16	0.449	0.46
17	0.614	0.602
18	0.92	0.917
19	0.244	0.184
20	1.0	1.0
total	0.8702	0.8799