Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Selectivity

Author: Bas Hamstra

Date: 22:10:26 01/12/01

Go up one level in this thread


On January 12, 2001 at 23:05:23, Christophe Theron wrote:

>On January 12, 2001 at 22:08:50, Bas Hamstra wrote:
>
>>I don't know if it useful to compare nodes this way. My first PV at depth=10
>>comes in the order of nodes of GT, but then it switches 4 times, to end at a4 at
>>22M nodes. I might be that GT was lucky to not have those expensive PV switches.
>>
>>Bas.
>
>
>
>It is true that you need to measure the number of nodes to reach a given ply
>depth on a number of typical positions before you can say which program is the
>most selective.
>
>However I can tell you in advance that you will (probably) discover that Chess
>Tiger and Gambit Tiger are extremely selective. And the next versions will be
>even more selective.
>
>
>
>    Christophe

No, no, don't do that! That's the way to promoting a pawn at move 216. Watching
grass grow is far more interesting. Do you really want to score 55.1% at SSDF in
stead on 54.93%? In stead of fireworks?

Slightly worried,
Bas.








>
>
>
>
>
>>On January 12, 2001 at 03:20:26, David Rasmussen wrote:
>>
>>>My program Chezzz, and most of the programs I've been learning from and looking
>>>at, has only little or moderate selectivity. Maybe I will keep it that way in
>>>Chezzz, but what are the possibilities?
>>>
>>>As far as I know, there are two general kinds of selectivity as currently used
>>>in "normal" alpha-beta based chess programs.
>>>
>>>1. Extensions
>>>2. Forward pruning
>>>
>>>I would like to know what hints and tricks you guys have about extensions,
>>>unusual extensions, how to limit extensions, when not to extend etc.
>>>
>>>But most of all, I would like to know about successful (I know this is somewhat
>>>subjective) forward pruning techniques, that are actually used in programs.
>>>
>>>I compared my programs performance on the position below, with what Christophe
>>>Theron posted from GT. At depth 10 the nodes searched by each programs are appx.
>>>
>>>GT     :  1.290.000
>>>Crafty :  13.500.000
>>>Chezzz :  12.500.000
>>>
>>>So Crafty and Chezzz is the same, and they are both very conservative in regard
>>>to selectivity IMO. I know that this is a choice and that one way is not better
>>>than the other, but I would still like to know how programs such as GT can get
>>>such a low node count.
>>>
>>>Now I know that GT is highly optimized, and that it has probably taken a long
>>>time get as "efficent" as GT. But I would still like to know as much as possible
>>>or at least the basics about how to forward prune succesfully like this.
>>>
>>>[D]2r1k2r/5pp1/4p3/ppqpP3/4bQPP/1B6/PPP2R1R/2K5 b k - 0 1



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.