🗄️ buffers

This repository is designed to provide two main features:

• easy bookkeeping of experience: the ReplayBuffer provides a simple interface to store and replay arbitrary experiences from single or parallelized environments while keeping episodes intact,
• efficient and scalable storage: experiences are stored in an EpisodeDataset, which builds on the PyTorch IterableDataset for efficient multi-process sampling.

Getting started

The quickstart notebook provides a brief overview of the main features of the package. Additional notebooks are available in the notebooks directory, demonstrating examples for multi-process sampling and composing of datasets.