https://github.com/bryanoliveira/gvgp-rl
A collection of reinforcement learning algorithms applied to General Video Game Playing.
https://github.com/bryanoliveira/gvgp-rl
Last synced: 8 months ago
JSON representation
A collection of reinforcement learning algorithms applied to General Video Game Playing.
- Host: GitHub
- URL: https://github.com/bryanoliveira/gvgp-rl
- Owner: bryanoliveira
- License: mit
- Created: 2019-04-24T01:24:13.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2019-07-08T23:52:13.000Z (over 6 years ago)
- Last Synced: 2025-01-03T01:54:45.472Z (10 months ago)
- Language: HTML
- Homepage:
- Size: 906 MB
- Stars: 1
- Watchers: 4
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# General Video Game Playing Reinforcement Learning Agents
A collection of reinforcement learning algorithms applied to General Video Game Playing. A good overview about the problem can be found [on this paper](https://arxiv.org/pdf/1802.10363.pdf).
## Requirements
- Python3
- Numpy
- Torch
- Gym
- Atari-py
- TensorboardX
- OpenCV 2
## Training
To quickly start training, run:
- `python3 main.py --game GAME_NAME --wrapper WRAPPER --model MODEL`
Or run `python3 main.py --help` to see all available options.
Example:
- `python3 main.py --game SpaceInvadersNoFrameskip-v0 --wrapper atari_conv --model a3c_conv`
To use Atari's image observation with `atari_conv` wrapper, GAME_NAME must contain `NoFrameskip` in the name.
## Testing
To test, you may use `--play --render` options:
- `python3 main.py --game GAME_NAME --wrapper WRAPPER --model MODEL --play --render`
Example:
- `python3 main.py --game SpaceInvadersNoFrameskip-v0 --wrapper atari_conv --model a3c_conv --play --render`
You can specify `--random` to run a random agent with the same configs and collect statistics. The `--render` option can be also specified on training to see Worker nº 0's performance.
## References
- [Rainbow: Combining Improvements in Deep Reinforcement Learning](https://arxiv.org/abs/1710.02298)
- [Asynchronous Advantage Actor-Critic (A3C)](https://arxiv.org/pdf/1602.01783.pdf)
- [Evolution Strategies](https://arxiv.org/pdf/1703.03864.pdf)