An offline deep reinforcement learning library

Get started API docs GitHub

The library providing professional-level deep reinforcement learning algorithms for everyone.

d3rlpy is the first library designed for not only researchers, but also practioners working on practical projects. This library provides both online and offline deep reinforcement learning algorithms with additional powerful features through the easy-to-use API.

features

OFFLINE RL ALGORITHMS

d3rlpy is the first to support offline deep reinforcement learning algorithms where the algorithm finds the good policy within the given dataset, which is suitable to tasks where online interaction is not feasible.

USER-FRIENDLY API

d3rlpy provides state-of-the-art algorithms through scikit-learn style APIs without compromising flexibility that provides detailed configurations for professional users. Moreoever, d3rlpy is not just designed like scikit-learn, but also fully compatible with scikit-learn utilites.

BEYOND STATE-OF-THE-ART

d3rlpy provides further tweeks such as ensemble algorithms and data augmentation to improve performance of state-of-the-art algorithms potentially beyond their original papers. Therefore, d3rlpy enables every user to achieve professional-level performance just in a few lines of codes.

quick start

            $ pip install d3rlpy

PyBullet

                  $ pip install git+https://github.com/takuseno/d4rl-pybullet
              
                
  import d3rlpy

  # prepare dataset
  dataset, _ = d3rlpy.datasets.get_pybullet('hopper-bullet-mixed-v0')
  # prepare algorithm
  cql = d3rlpy.algos.CQL(use_gpu=True)
  # start training
  cql.fit(dataset.episodes, n_epochs=100)

Atari 2600

                  $ pip install git+https://github.com/takuseno/d4rl-atari
              
                
  import d3rlpy

  # prepare dataset
  dataset, _ = d3rlpy.datasets.get_atari('breakout-mixed-v0')
  # prepare algorithm
  cql = d3rlpy.algos.DiscreteCQL(n_frames=4, scaler='pixel', use_gpu=True)
  # start training
  cql.fit(dataset.episodes, n_epochs=100)

video

family projects

d4rl-pybullet

Datasets for data-driven deep reinforcement learning with Pybullet environments

d4rl-pybullet is a dataset library providing continuous-control datasets collected with PyBullet environments.

GitHub

d4rl-atari

Datasets for data-driven deep reinforcement learning with Atari environments (wrapper for datasets released by Google)

d4rl-atari is a dataset library providing Atari datasets released by Google with convenience of automatic dataset management and easy-to-use API.

GitHub

MINERVA

An out-of-the-box GUI tool for offline deep reinforcement learning

MINERVA is a GUI tool for offline deep reinforcement learning without any coding.

GitHub