Author: Gerd Isenberg
Date: 23:23:21 03/02/06
Go up one level in this thread
On March 02, 2006 at 03:40:36, Volker Böhm wrote: >Hi Bob, > >for Spike I discovered the following: > >Statistically most fail highs occures at the first three moves. The fourth move >has again some fail highs, all others have very little fail highs. > >History based reduction gains at max. 20 Elo compared to a simple move ordering >based reduction (allways reduce after move 3). The move ordering is based on >history thus this is again some sort of history based pruning. > >Typically Spike plays better if more is reduced, I hardly found any >"less-reducing" rule that is better. Thus the test you are doing will show >little success for Spike as it will quickly drops beyond the "allways reduce" >level. > >In consequence the calculation of the history reduction threshold is very easy >in spike. Every time a move fails high it gets one point and every time a move >does not fail high (but a later move does) it is reduced by one point (there is >some depth related factor). Below zero: reduce; above zero: don´t reduce. >Surprisingly the value of a move does not grow too much and gets relative >stable. > >Greetings Volker So it seems that you already succesfully implemented Bob's idea with a 50% threshold. Thanks, Gerd
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.