https://github.com/twitter/diffusion-rl
https://github.com/twitter/diffusion-rl
Last synced: 6 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/twitter/diffusion-rl
- Owner: twitter
- License: apache-2.0
- Created: 2022-08-09T22:24:08.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2022-12-09T06:48:08.000Z (about 3 years ago)
- Last Synced: 2024-05-09T19:35:26.913Z (over 1 year ago)
- Language: Python
- Size: 35.2 KB
- Stars: 68
- Watchers: 5
- Forks: 15
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Diffusion-Offline-RL
In this work, we propose Diffusion-QL which utilizes a diffusion model as a highly expressive policy class for behavior cloning and policy regularization. In our approach we learn an action-value function and we add a term maximising action-values to the the training loss of the diffusion model, which results in a loss that seeks optimal actions that are near the behavior policy.
## Dependencies
Plese see the ``requirements.txt`` file for the detailed python package dependencies for our project.
## Run our Code
Running our code is quite easy, such as an example below,
```.bash
python run_offline.py --env_name walker2d-medium-expert-v2 --algo pcq
```