topological-equivariant-networks

Topological Equivariant Networks (TEN)

Setup instructions

git clone https://anonymous.4open.science/r/topological-equivariant-networks-853F/
cd topological-equivariant-networks
chmod +x env_builder.sh
source env_builder.sh

This will create and activate a new conda environment called "ten" that contains all the packages necessary to run the code in this repository.

To reproduce EGNN within the ETNN framework, run the following command:

TARGET_NAME=alpha LR=5e-4 source train_egnn_like.sh

To reproduce other target, set the corresponding TARGET_NAME.

The scripts to reproduce the reported QM9 experiments can be found under scripts/experiments. The experiments are named in the order of their appereance in the results table. For example, to reproduce the first row, run the following:

source scripts/experiments/1.sh 0 # creates the preprocessed dataset
source scripts/experiments/1.sh 1 # reproduces the results for the first 6 targets
source scripts/experiments/1.sh 2 # reproduces the results for the last 6 targets

The code to reproduce the spatial task is in the anonymized branch https://anonymous.4open.science/r/topological-equivariant-networks-27DB. The code to reproduce the synthetic tasks is in the anonymized branch https://anonymous.4open.science/r/topological-equivariant-networks-06C9. We plan to shortly merge these together under a single branch.