RANGE is a model-agnostic framework that employs an attention-based aggregation-broadcast mechanism to significantly reduce oversquashing effects in message-passing neural networks.
It is recommended to install the mlcg package (version 0.1.0) in a clean environment by following the official installation instructions available at https://github.com/ClementiGroup/mlcg.
This package provides the core infrastructure used to train and simulate all the models defined in this repository.
Install the rangemp package inside the newly created environment:
git clone git@github.com:ClementiGroup/range.git range
cd range; pip install .; cd ..