Author: Robert Hyatt
Date: 10:28:13 01/15/98
Go up one level in this thread
On January 15, 1998 at 13:25:20, Robert Hyatt wrote: >On January 15, 1998 at 10:23:54, Amir Ban wrote: > >> >>Hey, not fair ! You have the 50-move rule discount, not license to play >>around with the eval through the contempt factor. >> >>Evaluating the two passed pawns as worth the bishop would be wrong here. >>With so many other pawns and the black king positioned well black should >>in general win easily. This position is very unusual in that the white >>pawns are safe from attack while black's h7 cannot be protected. >> >>Amir > >Note that for CB I didn't touch anything. It evaluated the 50-move >draw as 0.000... I was suggesting dropping the contempt so that someone >could actually reach a position where a program might play Qd1, just to >see how it looks. > >In the case of CB, in a real game the draw score was never 0.000 anyway, >it was dynamic and depended on several things, somewhat like Crafty >although >not exactly. It would consider both clocks and think a draw was bad if >you >were behind. It would consider pieces on the board and king safety, and >if >the position is open enough, it would consider drawing bad. Etc. In >"problem" >mode however, this is disabled, so it used a pure 0.000... > >I did run this on Crafty, and it evaluates Qd1 followed by the sac as >-.4, >so I don't quite see why CB likes the position and Crafty doesn't (I let >crafty search to the same depth which is still pretty fast at 8 seconds >or >so) but I had to set the contempt to -1.000 and it found Qd1 quickly. >But >there is a .6 eval difference between the two programs. I haven't tried >to >find out why because I don't want to wade thru CB's assembly language to >see what it is doing. :) I should add that while CB would obviously solve this and win, if it were playing, there is a way where it might not be able to win. Since it is going to have to shuffle pieces for > 40 moves so that the 50-move rule would kick it in the right direction, it is possible that by the time 50 moves approaches, it can't find a way to play Qd1 without passing thru another position for the 3rd time and repeating. I'd hope there are enough paths, but whether they are is only one issue. The other is whether it would leave itself a path where Qd1 works, because the king has to be over on the q-side or it doesn't work. I'd call this 50-50...
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.