This repository contains the official implementation of our blogpost titled: Effect of equivariance on training dynamics.
Our blogpost has been published at GRaM Workshop @ ICML 2024.
Our extended blogpost can be found here.
To cite our work:
@misc{equivdyna_2024,
author = {Canez, Diego and Midavaine, Nesta and Stessen, Thijs and Fan, Jiapeng, Arias, Sebastian and Garcia, Alejandro},
doi = {10.5281/zenodo.14283519},
journal = {GRaM Workshop, ICML 2024},
month = jul,
title = {{Effect of equivariance on training dynamics}},
year = {2024}
}
To maximize convenience, reproducibility and encourage usage of our modules (models, datasets, tools), we've package some of them separately.
Make virtual environment and install dependencies:
make setup_envSource your virtual environment:
source .venv/bin/activateReproduce the training results from a given experiment:
python -m src.train experiment=wang2024/rgcnn_triple- ssh into snellius
- Move into the project root
cd ~/development/dl2
Let's say you want to run the experiment at configs/experiment/wang2022/equivariance_test/convnet.yaml. You can make use of the shortcut slurmtrain as follows:
make strain experiment=wang2022/equivariance_test/convnetIf you need to modify anything, the script is at scripts/slurm/train.sh.
make slurmcat id=6246500The logs are stored at scripts/slurm_logs/slurm_output_{id}.out.