README.md 1.34 KB
Newer Older
1
2
3
4
![No Maintenance Intended](https://img.shields.io/badge/No%20Maintenance%20Intended-%E2%9C%95-red.svg)
![TensorFlow Requirement: 1.x](https://img.shields.io/badge/TensorFlow%20Requirement-1.x-brightgreen)
![TensorFlow 2 Not Supported](https://img.shields.io/badge/TensorFlow%202%20Not%20Supported-%E2%9C%95-red.svg)

Devansh D's avatar
Devansh D committed
5
# Neural Programmer
Quoc Le's avatar
Quoc Le committed
6

kyscg's avatar
kyscg committed
7
Implementation of the Neural Programmer model as described in this [paper](https://openreview.net/pdf?id=ry2YOrcge).
Quoc Le's avatar
Quoc Le committed
8

kyscg's avatar
kyscg committed
9
10
11
12
Download and extract the data from the [WikiTableQuestions](https://ppasupat.github.io/WikiTableQuestions/) site. The dataset contains
11321, 2831, and 4344 examples for training, development, and testing respectively. We use their tokenization, number and date pre-processing. Please note that the above paper used the [initial release](https://github.com/ppasupat/WikiTableQuestions/releases/tag/v0.2) for training, development and testing. 

Change the `data_dir FLAG` to the location of the data.
Quoc Le's avatar
Quoc Le committed
13

Devansh D's avatar
Devansh D committed
14
### Training 
kyscg's avatar
kyscg committed
15
Run `python neural_programmer.py` 
Quoc Le's avatar
Quoc Le committed
16

kyscg's avatar
kyscg committed
17
The models are written to `FLAGS.output_dir`.
Devansh D's avatar
Devansh D committed
18
19

### Testing 
kyscg's avatar
kyscg committed
20
Run `python neural_programmer.py --evaluator_job=True`
Devansh D's avatar
Devansh D committed
21

kyscg's avatar
kyscg committed
22
The models are loaded from `FLAGS.output_dir`. The evaluation is done on development data.
Devansh D's avatar
Devansh D committed
23

kyscg's avatar
kyscg committed
24
In case of errors because of encoding, add `"# -*- coding: utf-8 -*-"` as the first line in `wiki_data.py`
Arvind Neekalantan's avatar
Arvind Neekalantan committed
25

Quoc Le's avatar
Quoc Le committed
26
Maintained by Arvind Neelakantan (arvind2505)