Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: History pruning

Author: Robert Hyatt

Date: 13:19:29 02/28/06

Go up one level in this thread


On February 28, 2006 at 14:48:10, Harald Lüßen wrote:

>On February 27, 2006 at 22:22:55, Robert Hyatt wrote:
>...
>>4.  the biggest thing I want to play with is the history values.  I am currently
>>"aging" the values between iterations, but what is happening is that the values
>>climb faster as the search gets deeper, which tends to make it get more
>>conservative as it goes deeper due to the history threshold limiting when
>>reductions are done.  Either the history values need to be stabilized, or
>>perhaps the reduction threshold needs to climb along with iteration number...
>>or something in between...
>
>When I read my history values I always get values between 0 (not good)
>and 256 (most often successfull). I do this with saved history values
>between 0 and 65536 and an additional maximum value which is always
>updated. Then I can return (history * 256 / (maximum + 1)). When maximum
>gets too big, all values and the maximum are divided by a factor.
>This is not very precise but you can easyly fix the flaws and
>find good tuning values for too big and factor.
>
>I do this because I want always be able to use the history
>as input value for some formulas like reductions or margins.
>Since my engine is not very good this might be a bad idea. ;-)
>
>Is it possible to make history pruning / late move reductions
>depend on the whole history value interval or is it important
>to use == 0 and != 0 only?
>
>Harald

I have a threshold value that can be changed.  0/!0 seems _very_ conservative,
whereas <N or >=N where N is greater than zero is more aggressive.  The larger
N, the more aggressive it is, but the more risky it is as well.

If history_value[x] >= N no reduction, else do reduction;  is the idea I am
using as one of the qualifying/limiting conditions...





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.