This is a Theano implementation of End-To-End Memory Networks, formerly known as Weakly Supervised Memory Networks. The dataset can be found here.
Below are the results obtained from the current implementation on the set of toy tasks.
MemN2N (Sukhbaatar 2015) | This Repo | |
---|---|---|
Task# | BoW | BoW |
1 | 0.6 | 37.6 |
2 | 17.6 | 49.1 |
3 | 71.0 | 77.8 |
4 | 32.0 | 31.4 |
5 | 18.3 | 23.5 |
6 | 8.7 | 27.9 |
7 | 23.5 | N/A |
8 | 11.4 | N/A |
9 | 21.1 | 19.5 |
10 | 22.8 | 26.1 |
11 | 4.1 | 30.1 |
12 | 0.3 | 52.2 |
13 | 10.5 | 46.0 |
14 | 1.3 | 21.8 |
15 | 24.3 | 0.0 |
16 | 52.0 | 55.5 |
17 | 45.4 | 49.1 |
18 | 48.1 | 55.7 |
19 | 89.7 | 81.4 |
20 | 0.1 | 0.0 |