Estimating search tree size with duplicate detection

Levi H.S. Lelis, Roni Stern, Nathan R. Sturtevant

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

5 Scopus citations

Abstract

In this paper we introduce Stratified Sampling with Duplicate Detection (SSDD), an algorithm for estimating the number of state expansions performed by heuristic search algorithms seeking solutions in state spaces represented by undirected graphs. SSDD is general and can be applied to estimate other state-space properties. We test SSDD on two tasks: (i) prediction of the number of A* expansions in a given f-layer when using a consistent heuristic function, and (ii) prediction of the state-space radius. SSDD has the asymptotic guarantee of producing perfect estimates in both tasks. Our empirical results show that in task (i) SSDD produces good estimates in all four domains tested, being in most cases orders of magnitude more accurate than a competing scheme, and in task (ii) SSDD quickly produces accurate estimates of the radii of the 4×4 Sliding-Tile Puzzle and the 3×3×3 Rubik’s Cube.

Original languageEnglish
Title of host publicationProceedings of the 7th Annual Symposium on Combinatorial Search, SoCS 2014
EditorsStefan Edelkamp, Roman Bartak
PublisherAAAI press
Pages114-122
Number of pages9
ISBN (Electronic)9781577356769
StatePublished - 1 Jan 2014
Event7th Annual Symposium on Combinatorial Search, SoCS 2014 - Prague, Czech Republic
Duration: 15 Aug 201417 Aug 2014

Publication series

NameProceedings of the 7th Annual Symposium on Combinatorial Search, SoCS 2014
Volume2014-January

Conference

Conference7th Annual Symposium on Combinatorial Search, SoCS 2014
Country/TerritoryCzech Republic
CityPrague
Period15/08/1417/08/14

ASJC Scopus subject areas

  • Computer Networks and Communications

Fingerprint

Dive into the research topics of 'Estimating search tree size with duplicate detection'. Together they form a unique fingerprint.

Cite this