NALU - Language to Number

Basic PyTorch implementation of Neural Arithmetic Logic Units by Andrew Trask, Felix Hill, Scott Reed, Jack Rae, Chris Dyer and Phil Blunsom.

• 4.1 Simple Function Learning Tasks
• 4.3 Language to Number Translation Tasks

Installation

git clone https://github.com/kerryeon/nalu or download

Requirements

1. Python >= 3.6
2. PyTorch >= 0.4
3. num2words from pip

Experiments

python ./test/simple_function.py
python ./test/language_to_number.py

Train TIP

Language to Number	Layer Size	Learning Rate	Plot
LSTM (en)	32	0.01
NALU (en)	32	0.01

Reference Source

Basic pytorch implementation from bharathgs : https://github.com/bharathgs/NALU

TODO

• More Languages
  • English
  • Korean
  • Japanese
  • Roman
  • Mayan
• Shuffle Dataset
• Furthermore
  • 4.2 MNIST Counting and Arithmetic Tasks
  • 4.4 Program Evaluation
  • 4.5 Track Time in a Grid-World Environment
• Simplify code
• Comments