NOTE: The repository is a work in progress.

Transformer Graph network for molecule property prediction. Uses 134k dataset to train for free energy, band gap, and atomic charges.

Graph-mol uses Makefile to perform training. Data is fetched and prepared using:

make prepare-data

Training is performed using:

make train model=graph_transformer

The partition for testing, training and validation, and seed for training is preserved for consistent training.

Jupyter notebooks to understand the Datamodules and the Graph model are present in notebooks. Jupyter lab can be run as:

make jupyter

The network can be deployed as a docker contatiner as:

make docker-build
make docker-run

To make contributions to the repository, check Contributing.md