A framework for training and evaluating Reinforcement Learning (RL) algorithms. Built with Python, TensorFlow framework and OpenAI Gym interface.
๐ง Currently under heavy development and some of its components may suffer from instability issues.
-
git clone https://github.com/dbobrenko/reinforceflow.git
-
cd reinforceflow
-
pip install -e .[tf-gpu]
In case if you have no CUDA device available, use CPU-only TensorFlow:
pip install -e .[tf]
-
To get examples working, install Gym by following the instructions at OpenAI Gym repo;
-
(Optional) For more environments, you can install OpenAI Universe;
Examples can be found in examples
directory:
# To train A3C on Breakout, see:
python examples/a3c/breakout.py
# To train Async DeepQ on Pong, see:
python examples/asyncdeepq/pong.py
# To train DQN on CartPole:
python examples/dqn/cartpole.py
# etc.
- DQN: Human-level control through deep reinforcement learning
- Double DQN: Deep Reinforcement Learning with Double Q-learning
- Dueling DQN Dueling Network Architectures for Deep Reinforcement Learning
- Prioritized Experience Replay
- Async DQN: Asynchronous Methods for Deep Reinforcement Learning
- A3C: Asynchronous Methods for Deep Reinforcement Learning