Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Searching PV first

Author: Stuart Cracraft

Date: 18:10:46 07/19/04

Go up one level in this thread


On July 19, 2004 at 18:21:26, Daniel Clausen wrote:

>On July 19, 2004 at 18:07:53, Stuart Cracraft wrote:
>
>>So I recently added some code to search the PV first
>>but with an add-on whose effect may be a curveball
>>in the longrun but appears okay at first.
>>
>>PV is searched first but with the add-on that any other
>>position that is searched (but not on the PV) there is a check
>>made for whether a PV-move at the same depth is available
>>in this position and if so it is searched first.
>>
>>Anyway, after doing the above my test results made another (small)
>>jump over just using the first move only of the prior iteration's
>>principal variation.
>>
>>Version 2.01
>>     ~Time  Solved Solved Time    Nodes     other information
>>    per move  %   / Total  Tot     Tot      other information
>>//   5 sec - 75% 227/300 1620.68 281651520 938838/5/173787 0/0/18410820/0/0/0
>>//   1 sec - 66% 199/300 323.91   53624628 178749/1/165555 0/0/3556160/0/0/0
>>
>>Version 2.00
>>//   5 sec - 73% 218/296 1687.07 301382976 1004610/6/178643 0 0/19507018/0/0/0
>>//   1 sec - 62% 187/300 313.89 52147824 173826/1/166134 0/0/3460861/0/0
>>
>>Stuart
>
>So will that become GNUchess6 in the end? :) If I may ask, do you start from
>scratch again or are you using something from GNUchessX?
>
>Sargon

It won't. GNU Chess will remain as Chua Kong-Sian's code from version 5
until something provably better comes along. For example, it has to
be readable and play reasonably well. That's a high bar. Beowulf and
Crafty strike me as two easy-to-obtain, better-written and better documented
chess programs than GNU Chess so I am not terribly motivated.

GA is just a toy to take up time. I had hoped to dabble in genetic
algorithms, hence the name, but got (predictably) caught up in all the
nitty-gritty which is gradually sorting itself out. At least it can
beat me easily now whereas a couple weeks ago it couldn't.

One person (Mridul) is working on pre-GNU-6 based on Vincent's move
generator -- but I think they both underestimate the value of bitboards
in the evaluation phase and the cost of evaluation in non-bitboard
programs, as stated by Bob. So I am not sure that such an approach will
bear fruit for the requirements that include readability and playability.

A bitboard evaluation is really easy to read and understand
for the most part. Other methods appear ugly and clumsy (and slower)
in comparison.

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.