ReinforcementLearningDatasets

A package to create, manage, store and retrieve datasets for Offline Reinforcement Learning. This package uses DataDeps.jl to fetch and track datasets.

Install

Since the package is not registered, you could install the package using the following command.

pkg> add https://github.com/JuliaReinforcementLearning/ReinforcementLearning.jl:src/ReinforcementLearningDatasets

Usage

Examples

D4RL dataset

using ReinforcementLearningDatasets
ds = dataset("hopper-medium-replay-v0"; repo="d4rl")
samples = Iterators.take!(ds)

ds is of the type D4RLDataset which consists of the entire dataset along with some other information about the dataset. samples are in the form of SARTS with batch_size 256.

RL Unplugged

using ReinforcementLearningDatasets
ds = rl_unplugged_atari_dataset("pong", 1, [1, 2])
samples = Iterators.take!(ds, 2)

ds is a Channel{RLTransition} that returns batches of type RLTransition when take! is used.

For more details refer to the documentation.

Note: The package is under active development and for now it supports only a limited number of datasets.

Supported Datasets

• D4RL: Datasets for Deep Data-Driven Reinforcement Learning
  • Mujoco datases and Pybullet datasets provided by D4RL are actively supported.
  • Flow and CARLA datasets have not been tested yet.
  • Mujoco Licence is not needed to access these Mujoco datasets but will be required for online evaluation.
• d4rl-pybullet
• Google Research atari DQN replay datasets
  • This directly loads the entire dataset into the RAM and should be used with caution. Features for multi threaded lazy loading will be provided soon.
• RL Unplugged: Benchmarks for Offline Reinforcement Learning
  • Currently supports atari datasets that are provided in RL Unplugged.
  • Multi threaded data loading is supported using a Channel that returns batches when Base.take! is used.