Author: Uri Blass
Date: 09:22:26 02/23/04
Go up one level in this thread
On February 23, 2004 at 12:15:17, Gerd Isenberg wrote: >On February 23, 2004 at 09:36:46, Uri Blass wrote: > >>On February 23, 2004 at 09:06:31, Gerd Isenberg wrote: >> >>>On February 23, 2004 at 07:02:59, martin fierz wrote: >>> >>>>aloha, >>>> >>>>i have a question about pins. pins are a rather important feature in chess; some >>>>of them are not so bad, some are deadly. i just happened to chat briefly with >>>>anthony cozzie on ICC, and he said he didn't do any pin detection. i detect >>>>pins, but i don't evaluate whether a pin is not so bad or deadly. my questions >>>>are: >>>>-> are you detecting pins in your program? >>>>-> if yes, do you try to distinguish between different pins? >>>> >>>>cheers >>>> martin >>> >>>Hi Martin, >>> >>>yes i detect pins, (not only) because i do legal move generation. >>>With disjoint direction attacks (from sliders as well from the king as >>>metasliders) it is rather cheap to get them without branches. >>> >>>In Eval i consider (from memory): >>> >>>1.) what kind of piece/pawn is pinned. >>>2.) whether the pinner (?) is en prise or attacked by equal valued pieces. >>>3.) The distance from pinned piece to the king (>2) and whether the pinned piece >>>is member of the "own" side of the board... >>>4.) whether the pinned piece is defended or defendable (in one move) by pawns. >>>5.) whether the piece is attackable by opposite pawns. >>>6.) a kind of SEE value considering all other attackers/defendes. >>> >>>In Eval i even consider other "tactical" stuff, like forks, overloading pieces >>>and pins to other valueable or hanging pieces. >>> >>>Cheers, >>>Gerd >> >>I wonder what is the reason that with all of this very impressive stuff you >>consider Yace and Sos as stronger than your program based on another post. > > >I don't know what Yace and Sos do in their eval - they were a bit unlucky at >IPCCC04. I found nothing impressive with considering pinned pieces and other >tactical stuff in eval. I believe that it is productive but not easy to implement correctly. Crafty has not this stuff and I believe that correct implementation of this stuff can help it. > > >> >>What is your opinion the relative advantages of program like Crafty relative to >>your program and I am not talking about parallel search because yace is not >>stronger than Crafty on one processor based on the results that I read. >> > >No idea. I don't do any autoplayer games. > > >>I wonder if you really test every change that you do in the evaluation to test >>if it is productive. >> >>I test every single change that I accept in the evaluation first in positions >>and later in blitz games(except cases that the change is relevant only to a very >>small class of positions) and if I do not get positive result in blitz games I >>reject the change. >> > >But may be a few combined changes leave a positive result, if each single change >does not. It is possible but I may try it only when I think that I have no simpler ideas to get improvement. Uri
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.