The Cross-environment Hyperparameter Setting Benchmark for Reinforcement Learning

This paper introduces a new empirical methodology, the Cross-environment Hyperparameter Setting Benchmark, that compares RL algorithms across environments using a single hyperparameter setting, encouraging algorithmic development which is insensitive to hyperparameters. We demonstrate that this benc...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Patterson, Andrew, Neumann, Samuel, Kumaraswamy, Raksha, White, Martha, White, Adam
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Patterson, Andrew
Neumann, Samuel
Kumaraswamy, Raksha
White, Martha
White, Adam
description This paper introduces a new empirical methodology, the Cross-environment Hyperparameter Setting Benchmark, that compares RL algorithms across environments using a single hyperparameter setting, encouraging algorithmic development which is insensitive to hyperparameters. We demonstrate that this benchmark is robust to statistical noise and obtains qualitatively similar results across repeated applications, even when using few samples. This robustness makes the benchmark computationally cheap to apply, allowing statistically sound insights at low cost. We demonstrate two example instantiations of the CHS, on a set of six small control environments (SC-CHS) and on the entire DM Control suite of 28 environments (DMC-CHS). Finally, to illustrate the applicability of the CHS to modern RL algorithms on challenging environments, we conduct a novel empirical study of an open question in the continuous control literature. We show, with high confidence, that there is no meaningful difference in performance between Ornstein-Uhlenbeck noise and uncorrelated Gaussian noise for exploration with the DDPG algorithm on the DMC-CHS.
doi_str_mv 10.48550/arxiv.2407.18840
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2407_18840</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2407_18840</sourcerecordid><originalsourceid>FETCH-arxiv_primary_2407_188403</originalsourceid><addsrcrecordid>eNqFjrEOgjAURbs4GPUDnOwPgEVL7CzRMLio7KQhD2m0j-a1IfL3VuLudJdz7z2MrTORSpXnYqvpbYZ0J8UhzZSSYs6uVQe8oN77BHAw1KMFDLwcHZDTpC0EIH6HEAw--BGw6aymJ2974jcwGLOBqXIBTRihJZu1-uVh9csF25xPVVEm03ntyMSBsf5K1JPE_j_xAadJPdI</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>The Cross-environment Hyperparameter Setting Benchmark for Reinforcement Learning</title><source>arXiv.org</source><creator>Patterson, Andrew ; Neumann, Samuel ; Kumaraswamy, Raksha ; White, Martha ; White, Adam</creator><creatorcontrib>Patterson, Andrew ; Neumann, Samuel ; Kumaraswamy, Raksha ; White, Martha ; White, Adam</creatorcontrib><description>This paper introduces a new empirical methodology, the Cross-environment Hyperparameter Setting Benchmark, that compares RL algorithms across environments using a single hyperparameter setting, encouraging algorithmic development which is insensitive to hyperparameters. We demonstrate that this benchmark is robust to statistical noise and obtains qualitatively similar results across repeated applications, even when using few samples. This robustness makes the benchmark computationally cheap to apply, allowing statistically sound insights at low cost. We demonstrate two example instantiations of the CHS, on a set of six small control environments (SC-CHS) and on the entire DM Control suite of 28 environments (DMC-CHS). Finally, to illustrate the applicability of the CHS to modern RL algorithms on challenging environments, we conduct a novel empirical study of an open question in the continuous control literature. We show, with high confidence, that there is no meaningful difference in performance between Ornstein-Uhlenbeck noise and uncorrelated Gaussian noise for exploration with the DDPG algorithm on the DMC-CHS.</description><identifier>DOI: 10.48550/arxiv.2407.18840</identifier><language>eng</language><subject>Computer Science - Learning</subject><creationdate>2024-07</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,777,882</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2407.18840$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2407.18840$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Patterson, Andrew</creatorcontrib><creatorcontrib>Neumann, Samuel</creatorcontrib><creatorcontrib>Kumaraswamy, Raksha</creatorcontrib><creatorcontrib>White, Martha</creatorcontrib><creatorcontrib>White, Adam</creatorcontrib><title>The Cross-environment Hyperparameter Setting Benchmark for Reinforcement Learning</title><description>This paper introduces a new empirical methodology, the Cross-environment Hyperparameter Setting Benchmark, that compares RL algorithms across environments using a single hyperparameter setting, encouraging algorithmic development which is insensitive to hyperparameters. We demonstrate that this benchmark is robust to statistical noise and obtains qualitatively similar results across repeated applications, even when using few samples. This robustness makes the benchmark computationally cheap to apply, allowing statistically sound insights at low cost. We demonstrate two example instantiations of the CHS, on a set of six small control environments (SC-CHS) and on the entire DM Control suite of 28 environments (DMC-CHS). Finally, to illustrate the applicability of the CHS to modern RL algorithms on challenging environments, we conduct a novel empirical study of an open question in the continuous control literature. We show, with high confidence, that there is no meaningful difference in performance between Ornstein-Uhlenbeck noise and uncorrelated Gaussian noise for exploration with the DDPG algorithm on the DMC-CHS.</description><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNqFjrEOgjAURbs4GPUDnOwPgEVL7CzRMLio7KQhD2m0j-a1IfL3VuLudJdz7z2MrTORSpXnYqvpbYZ0J8UhzZSSYs6uVQe8oN77BHAw1KMFDLwcHZDTpC0EIH6HEAw--BGw6aymJ2974jcwGLOBqXIBTRihJZu1-uVh9csF25xPVVEm03ntyMSBsf5K1JPE_j_xAadJPdI</recordid><startdate>20240726</startdate><enddate>20240726</enddate><creator>Patterson, Andrew</creator><creator>Neumann, Samuel</creator><creator>Kumaraswamy, Raksha</creator><creator>White, Martha</creator><creator>White, Adam</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20240726</creationdate><title>The Cross-environment Hyperparameter Setting Benchmark for Reinforcement Learning</title><author>Patterson, Andrew ; Neumann, Samuel ; Kumaraswamy, Raksha ; White, Martha ; White, Adam</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-arxiv_primary_2407_188403</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Patterson, Andrew</creatorcontrib><creatorcontrib>Neumann, Samuel</creatorcontrib><creatorcontrib>Kumaraswamy, Raksha</creatorcontrib><creatorcontrib>White, Martha</creatorcontrib><creatorcontrib>White, Adam</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Patterson, Andrew</au><au>Neumann, Samuel</au><au>Kumaraswamy, Raksha</au><au>White, Martha</au><au>White, Adam</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>The Cross-environment Hyperparameter Setting Benchmark for Reinforcement Learning</atitle><date>2024-07-26</date><risdate>2024</risdate><abstract>This paper introduces a new empirical methodology, the Cross-environment Hyperparameter Setting Benchmark, that compares RL algorithms across environments using a single hyperparameter setting, encouraging algorithmic development which is insensitive to hyperparameters. We demonstrate that this benchmark is robust to statistical noise and obtains qualitatively similar results across repeated applications, even when using few samples. This robustness makes the benchmark computationally cheap to apply, allowing statistically sound insights at low cost. We demonstrate two example instantiations of the CHS, on a set of six small control environments (SC-CHS) and on the entire DM Control suite of 28 environments (DMC-CHS). Finally, to illustrate the applicability of the CHS to modern RL algorithms on challenging environments, we conduct a novel empirical study of an open question in the continuous control literature. We show, with high confidence, that there is no meaningful difference in performance between Ornstein-Uhlenbeck noise and uncorrelated Gaussian noise for exploration with the DDPG algorithm on the DMC-CHS.</abstract><doi>10.48550/arxiv.2407.18840</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2407.18840
ispartof
issn
language eng
recordid cdi_arxiv_primary_2407_18840
source arXiv.org
subjects Computer Science - Learning
title The Cross-environment Hyperparameter Setting Benchmark for Reinforcement Learning
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-20T04%3A51%3A32IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=The%20Cross-environment%20Hyperparameter%20Setting%20Benchmark%20for%20Reinforcement%20Learning&rft.au=Patterson,%20Andrew&rft.date=2024-07-26&rft_id=info:doi/10.48550/arxiv.2407.18840&rft_dat=%3Carxiv_GOX%3E2407_18840%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true