We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Tianshou: a Highly Modularized Deep Reinforcement Learning Library

Abstract: We present Tianshou, a highly modularized python library for deep reinforcement learning (DRL) that uses PyTorch as its backend. Tianshou provides a flexible, reliable, yet simple implementation of a modular DRL library, and has supported more than 20 classic algorithms succinctly through a unified interface. To facilitate related research and prove Tianshou's reliability, we have released Tianshou's benchmark on MuJoCo environments, covering eight classic algorithms with the state-of-the-art performance. We open-sourced Tianshou at this https URL, which has received over 3.7k stars and become one of the most popular PyTorch-based DRL libraries.
Comments: 10 pages, 4 figures, 4 tables
Subjects: Machine Learning (cs.LG)
Cite as: arXiv:2107.14171 [cs.LG]
  (or arXiv:2107.14171v2 [cs.LG] for this version)

Submission history

From: Huayu Chen [view email]
[v1] Thu, 29 Jul 2021 16:49:03 GMT (1115kb,D)
[v2] Wed, 22 Sep 2021 07:17:45 GMT (1228kb,D)

Link back to: arXiv, form interface, contact.