Generalized Population-Based Training With Pairwise Learning (GPBT-PL) arXiv

Code for the Generalized Population-Based Training With Pairwise Learning (GPBT-PL) algorithm, from the paper Generalized Population-Based Training for Hyperparameter Optimization in Reinforcement Learning.

The GPBT framework is implemented based on ray. Heavily inspired by ray tune PBT example, GPBT-PL is included in the ray.tune library, which is the official supported implementation.

Running the code

To run the PPO experiment, use command:

python run_ppo.py 

To run the IMPALA experiment, use command:

python run_impala.py

Citing GPBT-PL

@article{bai2024generalized,
  title={Generalized Population-Based Training for Hyperparameter Optimization in Reinforcement Learning}, 
  author={Hui Bai and Ran Cheng},
  journal={IEEE Transactions on Emerging Topics in Computational Intelligence},
  publisher = {IEEE},
  year={2024},
  doi={10.1109/TETCI.2024.3389777}
  }