Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
gce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Reproduction of 'Analysing Mathematical Reasoning Abilities of Neural Models' Saxton et. al. 2019

Setup

  1. Clone the repo

  2. Set up a Python 3.7 environment

  3. Enter the repo folder and run pip install -r requirements.txt

  4. Optional: download the full 2GB dataset (https://console.cloud.google.com/storage/browser/_details/mathematics-dataset/mathematics_dataset-v1.0.tar.gz) and unzip in the project root directory.

Usage

If you're running on a non CUDA machine (i.e. a laptop), these commands will run with reduced batch size and dataset size for faster testing/debugging.

Training

To train a model you must select from Vanilla Transformer, Simple LSTM, and Attentional LSTM. Options transformer, simLSTM, and attLSTM, respectively.

For example: python training.py -m transformer

Benchmarking

To run performance benchmarks on the Transformer run: python benchmark.py

Visualization

Tensorboard logs are saved to the runs folder.

tensorboard --logdir runs

Colaborators

  • Andrew Schreiber
  • Taylor Kulp-McDowall

Links

Relevant papers

About

Reproduction of 'Analysing Mathematical Reasoning Abilities of Neural Models' Saxton et. al. 2019

Resources

Packages

No packages published