Author: martin fierz
Date: 14:30:45 08/03/01
Go up one level in this thread
On August 03, 2001 at 16:56:08, Artem Pyatakov wrote: >Has anyone here had any luck with Enhanced Transposition Cutoffs in their >program? > >Just out of curiosity (and ease of SLOW/SIMPLE implementation of this idea), I >decided to try it. > >The way I implemented it (just as a test), is by increasing move ordering >"priorityScore" for successor moves that promise a cutoff and thus trying these >moves first (but after the hash move). But for some strange reason, I can't >even get it to marginally reduce the number of nodes. > >This does not make sense to me, since I think that: >1) It's supposed to reduce the number of nodes at least somewhat >2) The reason most do not use is because the number of nodes that it eliminates >trades off with the computational time, NOT because this approach does not work >completely. > >Any ideas? suggestions? > >Thank you. > >Artem I am using ETC in my checkers program (much smaller branching factor than chess), and it is useful. i guess it reduces my nodes by roughly 20% on a 1-minute search, but it also causes about 10% overhead. the net result is positive though. my implementation is more or less straight from schaeffers paper on the subject. it's important that you use ETC only in nodes which are far enough away from the leaf nodes, i am using it only if there are >= 2 ply remaining to be searched. an ETC lookup needs some time, and only if the gain is large enough it makes sense to do it. i am not using it to order moves but to make immediate cutoffs - that's the idea behind ETC. cheers martin
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.