Awesome

Better Training of GFlowNets with Local Credit and Incomplete Trajectories

This repository is the implementation of Better Training of GFlowNets with Local Credit and Incomplete Trajectories in ICML 2023. This codebase is based on the open-source gflownet implementation, and please refer to that repo for more documentation.

Citing

If you used this code in your research or found it helpful, please consider citing our paper:

@inproceedings{
	pan2023better,
	title={Better Training of GFlowNets with Local Credit and Incomplete Trajectories},
	author={Ling Pan and Nikolay Malkin and Dinghuai Zhang and Yoshua Bengio},
	booktitle={International Conference on Machine Learning},
	year={2023},
	url={https://openreview.net/forum?id=beHp3L9KXc}
}

Requirements

Grid

• python: 3.6
• torch: 1.3.0
• scipy: 1.5.4
• numpy: 1.19.5
• tdqm

Molecule discovery

Please check the gflownet repo for more details about the environment

Usage

Please follow the instructions below to replicate the results in the paper.

• Grid

python gflownet.py --method <METHOD> --fl <FL_FLAG> --size <SIZE> --seed <SEED>

Specifiy METHOD=db_gfn with FL_FLAG=1 for FL-GFN.

• Molecule discovery

python gflownet.py --objective <OBJECTIVE> --fl <FL_FLAG> --run <RUN>