Awesome

BatchGFN

BatchGFN is a batch active learning method which uses a generative flow network trained to sample batches of points from a pool dataset proportional to a batch reward function.

The code in this repository was used in the ICML 2023 SPIGM workshop paper BatchGFN: Generative Flow Networks for Batch Active Learning, Shreshth A. Malik, Salem Lahlou, Andrew Jesson, Moksh Jain, Nikolay Malkin, Tristan Deleu, Yoshua Bengio, Yarin Gal. (Arxiv).

Installation

git clone git@github.com:s-a-malik/batchgfn.git
cd batchgfn
conda env create -f environment.yml
conda activate batchgfn

If it fails you might need to

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source $HOME/.cargo/env

Usage

Use command line arguments to set experimental variables. scripts/main.py is the main entry point for active learning experiments.

Configs and logs are saved to wandb. Run wandb login before running, and wandb offline if you do not want to save logs online.

Cite

If you found our work useful, please consider citing:

@article{malik2023batchgfn,
  title={BatchGFN: Generative Flow Networks for Batch Active Learning},
  author={Malik, Shreshth A and Lahlou, Salem and Jesson, Andrew and Jain, Moksh and Malkin, Nikolay and Deleu, Tristan and Bengio, Yoshua and Gal, Yarin},
  journal={arXiv preprint arXiv:2306.15058},
  year={2023}
}

Disclaimer

This is research code shared without support or guarantee of quality. Please let us know, however, if there is anything wrong or that could be improved and we will try to solve it.