zunction / relation-network-1

Tensorflow implementation of Relation Network (bAbI dataset)

A simple neural network module for relational reasoning

paper link: https://arxiv.org/abs/1706.01427

Tensorflow implementation of Relation Network on bAbI dataset

Result

Task	Accuracy	Success/Fail
Task 1	0.9990	S
Task 2	0.8422	F
Task 3	0.8020	F
Task 4	1.0000	S
Task 5	0.9920	S
Task 6	0.9960	S
Task 7	0.9869	S
Task 8	0.9719	S
Task 9	0.9980	S
Task 10	0.9899	S
Task 11	0.9940	S
Task 12	0.9990	S
Task 13	1.0000	S
Task 14	0.9970	S
Task 15	1.0000	S
Task 16	0.4324	F
Task 17	0.8760	F
Task 18	0.9669	S
Task 19	0.6026	F
Task 20	1.0000	S

total: 15/20

Prerequisites

• Python 3.5+
• Tensorflow 1.0.1
• Numpy
• argparse
• itertools
• os
• pickle
• re
• sys
• datetime
• time

Usage

1. Load data

$ python preprocessing.py --path 'path-where-tasks_1-20_v1-2-located'

2. Run model

$ python train.py