Author: leonid
Date: 03:31:19 07/25/00
Go up one level in this thread
On July 24, 2000 at 19:16:36, Eelco de Groot wrote: > >Yes Leonid, that was exactly what I saw too, at least in Rebel Century 1.2! >Almost no change from 28 MB to 0 kB.. That's why I thought that positions found >in the hastables maybe generally don't get counted. But it seems I was wrong >again. Anyway to measure the speed of a program it is maybe better then to try >out some positions with some key-moves that you definitely want to see your >program play and measure the time your program takes to find them. Fast times >are not yet a guarantee of good play throughout all stages of a game but at >least testing can be done in reasonable time. Another way of testing that I >could think of at later stages of development might be to find move sequences, >from GM games for example, that go together and that you think would be just at >the edge of what your program can do. Because, if your program misses one of the >moves, the other moves then don't make any sense anymore to play even if they >were objectively very good. If the program had a way of sensing that the >position was becoming critical in such a sequence, more pieces being attacked or >more early cutoffs found for bad moves maybe, -I'm thinking this part up on the >spot-, it could spend more time thinking on its move, ideally. > >Sorry for adding to the confusion! > >Eelco I am trying to do something like you said. I do try to see if something is wrong by detecting wrong move in brute force search in positions that have one possible good move. Other way around is comparing best brute force move with minimax search that is perfect but slow like hell. This verification must be done all the time since after each speeding of program new bugs are all the time possible. Thanks for responding anyway! Confusion is impossible when idea is clear and verifiable. Myself I jumped already many times from one idea to the opposit one while writing my program. The last speeding was done exactly when I found that one saying by Karrigan and Bob H. could have some sense even for my program. It was said that producing all legal moves for each ply is a waist of time, since only few will be used. This sound to me as something slightly wrong because you can make your choice of moves only when you have them all before you. But recently I looked how many times program find best move with just one move. Number was above 50%. So, now my program produce best move first (when it existe) and if this move will not deliver final result on the spot, goes to generate all legal moves and making search usual way. This make jump my NPS and speed in very merry way. And how many old wrong ideas was still not proved to be perfectly right! Leonid.
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.