Author: Andrew Dados
Date: 22:31:07 04/16/00
Go up one level in this thread
At first I need to notice I never saw any commercial programmer post on topic of move ordering...:) Now my few loose remarks: I didn't find history heuristic profitable at all, maybe because I already use some forms of 'dynamic' move ordering. I try to get ordering data from search rather then history. Here's example: White on previous ply moved Pe2-e4, threatening PxNd5 (which I don't know yet). Now black null-moves. Return from null move is <alpha, best move on next ply is capture of Nd5. I (as black) try killers, captures, when I get to 'rest of moves' section (part which gets ordered by history heuristic), I start with 'escape' moves for Nd5. In middle game I get 45-60% of 'other move cutoffs' from those 'escape' moves. Moreover, if such escape move gives me a cutoff, I don't update killers with it (I treat is as 'special move', as check evasions). There is more tricks you may find around such and similar concepts here... I wonder if someone can provide data on % cutoffs given by history heutistics versus total 'all other moves' cutoffs. Say for 3 first history moves. In most endgames I find now 'response killers' to work nice for my program (previous non-capture response which gave cutoff to opponents move). As usual, YMMV... but experimenting with move ordering other then 'what all do' can be definitely fun and profitable. (check nodecounts for Diep or Fritz for example...:) -Andrew-
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.