Welcome to my personal project on reinforcement learning! This repository contains solutions to the Gymnasium problem set, using deep reinforcement learning and Q-learning algorithms.
The code in this repository is written in Python and includes both single-threaded and multi-threaded implementations. I created this project to learn about reinforcement learning and see its capabilities firsthand. With this repository, you can train your own agents and experiment with deep reinforcement learning and Q-learning on a variety of environments, including Atari games, classic-control, and toy-text.
- Explore deep reinforcement learning and Q-learning algorithms and their performance on various environments
- Train agents to solve complex Atari games using multi-threading
- Use the code as a starting point for your own reinforcement learning projects
- Nvidia GPU with CUDA cores
- Graphics driver
- CUDA
- cuDNN
The best way to obtain 2, 3, and 4 is to follow the instructions provided by Nvidia.
Latest driver version should be compatible with older CUDA versions. Graphics drivers are supposed to be backwards compatible with CUDA versions.
The CUDA version chosen seems to have a dependancy on OS version.
Make sure to install compatible version combinations for Tensorflow.
To verify install and check versions of graphics driver and CUDA run:
nvidia-smiInstall Tensorflow.
pip install tensorflowInstall Gymnasium.
pip install gymnasium gymnasium[atari] gymnasium[accept-rom-license]Install additional Python library requirements.
pip install opencv-python numpy matplotlib joblibI welcome contributions to this repository! If you have any feedback, suggestions, or ideas for improvements, please feel free to open an issue or submit a pull request.