This repository consists of a basic framework that allows to easily load a Gym environment and optimize an agent using SAC. The SAC algorithm: Soft-Actor-Critic.
The code is written in python 3.10 and uses the following packages:
Experiments with a set of hyperparameters can be run with the main.py
script.
Experiments to find the best hyperparameters can be run with the hpo.py
script.
The evaluate.py
script can be used to evaluate a trained agent.
- Make a version that uses PyTorch Lightning
- With PyTorch Lightning, make use of Weights & Biases for logging
- Add support for distributed training (Vertex AI)