Author: Stuart Cracraft
Date: 11:25:55 01/03/06
Go up one level in this thread
On January 03, 2006 at 14:09:14, Uri Blass wrote: >On January 03, 2006 at 14:04:17, Stuart Cracraft wrote: > >>On January 03, 2006 at 12:53:33, Uri Blass wrote: >> >>>On January 03, 2006 at 12:18:58, Tord Romstad wrote: >>> >>>>On January 03, 2006 at 11:49:05, Robert Allgeuer wrote: >>>> >>>>>On January 03, 2006 at 10:49:54, Maurizio Monge wrote: >>>>> >>>>>>What you said is surely true. >>>>>>But what i find strange is that, IIRC, the only quite new technic in computer >>>>>>chess that can be found in fruit is history pruning, everything else is just a >>>>> >>>>>History pruning was already in use in SmarThink and other engines before as >>>>>well. If I am not completely mistaken history pruning was invented by Sergej for >>>>>SmarThink. >>>> >>>>It is possible that Sergei introduced the name "history pruning", but the >>>>technique itself is very old; certainly much older than SmarThink. >>> >>>You are right and Movei use it for some years. >>>First public version of movei to use history based reduction was 07_99 >>> >>>I did not talk about it at that time but I used it and I am not going to be >>>surprised if other also used it earlier. >>> >>> I no >>>>longer remember where or when I heard about it for the first time, but it was >>>>definitely not in this millennium. >>>> >>>>"History pruning" is a really bad name for the technique, by the way. Since >>>>a long time, I have been advocating to rename it to "late move reductions". >>>> >>>>The word "history" is misleading because the technique can be implemented >>>>without using history counters. >>> >>>In this case it is not history based pruning. >>>I certainly use history counters but it is possible that I can improve it by not >>>using history counters and using different conditions instead of them. >>> >>>Today I use combination of evaluation and history information to decide about >>>reduction. >>> >>> >>> I currently use a combination of null move >>>>threat detection and evaluation data to make my late move reduction decisions, >>>>and don't use history counters at all. This seems to work clearly better, >>>>at least in my program. >>> >>>originally when I implemented it I had no condition about late move reduction >>>but later I changed it and at least today I never reduce the first move. >>> >>>> >>>>The word "pruning" is misleading because most people don't use the idea >>>>to prune moves, but only to reduce the search depth. >>> >>>I agree that the word reduction is better. >>>> >>>>"Late move reductions" is a much more appropriate name, and does a better >>>>job of explaining what the idea is about: Reducing the depth for the less >>>>interesting moves late in the move list. >>>> >>>>Tord >>> >>>The idea is to reduce the depth of moves that you are almost sure that they are >>>going to fail low. >>> >>>Uri >> >>And how do you decide this? > >You can decide about it based on combination of evaluation and history of the >search and the place of the move in the move list. > >The implementation is different in different programs. > >Uri Hmm - sounds rather free-form. Let me know if you have an published articles on it. I'd like to read about it. Thanks, 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.