DMCA
Experimental Studies of Variable Selection Strategies Based on Constraint Weights
| Citations: | 7 - 3 self |
BibTeX
@MISC{Wallace_experimentalstudies,
author = {Richard J. Wallace and Diarmuid Grimes},
title = {Experimental Studies of Variable Selection Strategies Based on Constraint Weights},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Abstract. An important class of heuristics for constraint satisfaction problems works by sampling information during search in order to inform subsequent decisions. One of these strategies, called the “weighted degree heuristic”, is based on weighting constraints according to their involvement in failure during search. Recently, a new approach to sampling based on weighted degree was introduced, that uses a form of “random probing ” to gain information that is less subject to sampling bias. This approach also involves restarting in order to enhance the initial choices made during search. The present research analyses the characteristics of the sampling process and the manner in which information is used, to better understand strategies based on constraint weights. Using a framework based on two performance principles, the well-known Fail-First and a “Contention Principle”, we study the properties of both the sampling and variable selection components. We show that the critical factor in search enhancement is improved fail-firstness (rather than initial choices), that the important concept wrt sampling is contention rather than failure, that probing effectiveness depends on the quality of the sample, and that random probing seems to balance two basic forms of heuristic action to improve overall search efficiency. 1
Keyphrases
constraint weight experimental study variable selection strategy initial choice contention principle well-known fail-first search enhancement new approach sampling process understand strategy weighted degree heuristic important class important concept wrt sampling basic form performance principle overall search efficiency constraint satisfaction problem work heuristic action subsequent decision effectiveness depends weighted degree critical factor present research variable selection component
