Understand the project
Practical overview
When it fits
Use it when your reinforcement-learning work is based on Python and Keras and you want included algorithms or extension points for custom environments and agents.
Who it is for
- Python developers building reinforcement-learning experiments with Keras.
- Researchers and developers comparing supported algorithms in Gym environments.
- Developers extending agents, environments or algorithms for their own projects.
Documented uses
- Build and evaluate agents using included deep reinforcement-learning algorithms.
- Prototype game-playing policies in environments such as CartPole.
- Add custom environments, algorithms, callbacks or metrics to a Keras-based workflow.
- Log training metrics through the documented optional Weights & Biases callback.
Before you choose
Requirements & limits
The project is available as a public GitHub repository with an MIT license. Its README recommends installation from PyPI with pip.
- The README lists DQN, Double DQN, DDPG, Continuous DQN (NAF), Cross-Entropy Method, Dueling DQN and Deep SARSA as implemented.
- A3C and PPO are explicitly shown as not implemented in the README.
- Running examples requires Gym and h5py in addition to the library; Atari examples also call for Pillow and Gym’s Atari module.
- Training-metric visualization with Weights & Biases is an optional callback path and requires installing wandb.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Source-based practical guide
Install dependencies and start the CartPole example
Follow the README’s documented setup and command for its DQN CartPole example.
Before you start
- Keras Rl installed.
- Gym and h5py installed for examples.
Steps
Install Gym following the Gym installation instructions.
Install h5py with the command below.
pip install h5pyRun the command below.
python examples/dqn_cartpole.py
Expected outcome
The documented CartPole DQN example starts; the README says it visualizes the game during training.
Guide limitations
- The instructions do not provide a complete environment-specific dependency or compatibility matrix.
Evidence and transparency
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- keras-rl/keras-rl · Official homepage title · retained in its source language