Huskarl is a framework for deep reinforcement learning focused on research and fast prototyping. It's built on TensorFlow 2.0 and uses the tf.keras API when possible for conciseness and readability.

Huskarl makes it easy to parallelize computation of environment dynamics across multiple CPUs. This is useful for speeding up on-policy learning algorithms that benefit from multiple concurrent sources of experience such as A2C or PPO. It is specially useful for computationally intensive environments such as physics-based ones.

Huskarl works seamlessly with OpenAI Gym environments.

There are plans to support multi-agent environments and Unity3D environments.

Several algorithms are implemented already and many more are planned.

• Deep Q-Learning Network (DQN)
• Multi-step DQN
• Double DQN
• Dueling Architecture DQN
• Advantage Actor-Critic (A2C)
• Deep Deterministic Policy Gradient (DDPG)
• Proximal Policy Optimization (PPO)
• Prioritized Experience Replay
• Curiosity-Driven Exploration

Since TensorFlow 2.0 is not officially out yet you need to install it and other dependencies manually for now:

pip install tf-nightly-2.0-preview
pip install cloudpickle
pip install scipy
pip install huskarl --no-deps

If you use Huskarl in your research, you can cite it as follows:

@misc{salvadori2019huskarl,
    author = {Daniel Salvadori},
    title = {huskarl},
    year = {2019},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/danaugrs/huskarl}},
}

hùskarl in Old Norse means a warrior who works in his/her lord's service.