A simple neural network module for relational reasoning
paper link: https://papers.nips.cc/paper/7082-a-simple-neural-network-module-for-relational-reasoning.pdf
Tensorflow implementation of Relation Network on bAbI dataset
Result
| Accuracy | Success / Fail | |
|---|---|---|
| Task 1 | 1 | S |
| Task 2 | 0.935 | F |
| Task 3 | 0.871 | F |
| Task 4 | 1 | S |
| Task 5 | 0.995 | S |
| Task 6 | 1 | S |
| Task 7 | 0.998 | S |
| Task 8 | 0.999 | S |
| Task 9 | 1 | S |
| Task 10 | 1 | S |
| Task 11 | 0.996 | S |
| Task 12 | 1 | S |
| Task 13 | 1 | S |
| Task 14 | 1 | S |
| Task 15 | 1 | S |
| Task 16 | 0.497 | F |
| Task 17 | 0.991 | S |
| Task 18 | 0.994 | S |
| Task 19 | 0.979 | S |
| Task 20 | 1 | S |
total: 17/20
Prerequisites
- Python 3.5+
- Tensorflow-gpu 1.1+
- Numpy
- argparse
- itertools
- os
- pickle
- re
- sys
- datetime
- time
Usage
- Load data
$ python preprocessing.py --path 'path-where-tasks_1-20_v1-2-located'
- Run model
$ python train.py