Author: Stuart Cracraft
Date: 14:53:07 09/04/04
Go up one level in this thread
On September 04, 2004 at 11:36:24, Dan Honeycutt wrote: >On September 03, 2004 at 10:29:11, Stuart Cracraft wrote: > >> >>I am glad you raised this as I wanted to talk about. My program is a mixture >>of return(value), return(beta), and return(alpha) and I've not been happy with >>that. My quiescence is mostly return(beta). The main search is a mixture. >>Whenever I make everything return(value), PVS goes weird on me and stops >>solving problems as well (big drop). What should I do? >> > >Hi Stuart: > >I think its a function of what you store in the hash. Say you get back a value >> beta. You are going to cut and save a lower bound score in the hash. If the >position occurs again you are going to get a hash cut if the stored score is >greater than some future beta. If you stored the higher score (value instead of >beta) your chances for a hash cut are greater. Thus less nodes but also less >vision. > >I use fail soft. I get more depth and worse results on test suites but it seems >to play better. YMMV. > >Dan H. Don't worry. I sorted this out on Friday. Uri's goading helped. I am now a completely return-value type of program with no impact to test results. I did notice changes in the speed of the search and number of moves searched (both higher) when I changed from a mixture of return value (potentially a newly-set alpha) and return beta to all return value. But glad to have it all that way -- I was hoping that it would help getting the Botvinnik-Markoff extension or Mate Threat extension showing some good results but not so lucky. Mate Threat for me is still just par, no change across the suite and neither helps me on WAC 141 though having recapture solves it at least. Stuart
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.