Author: Robert Hyatt
Date: 20:23:09 05/10/02
Go up one level in this thread
On May 10, 2002 at 19:21:13, Marc van Hal wrote: >On May 09, 2002 at 20:29:37, Robert Hyatt wrote: > >>On May 09, 2002 at 18:08:52, J. Wesley Cleveland wrote: >> >>>On May 09, 2002 at 13:22:01, Robert Hyatt wrote: >>> >>>>On May 09, 2002 at 12:43:08, Robert Hyatt wrote: >>>> >>>>>On May 08, 2002 at 16:22:16, J. Wesley Cleveland wrote: >>>>> >>>>>>In this position which is an obvious draw for any decent human player, >>>>>>[D]8/8/p2b4/8/p7/3k4/1B6/1K6 w - - 0 4 >>>>>>crafty is clueless - static evaluation is -3.19, evaluation after 34 ply is >>>>>>still -3.19. >>>>> >>>>> >>>>>This is a harder case that my current "knowledge" doesn't cover. Black >>>>>has the right bishop for its rook pawns. Of course white has the same >>>>>colored bishop and if they are traded, it is a draw because of simple >>>>>rook pawns left with the opposing king controlling the queening square. >>>>> >>>>>I can fix this cheaply and will. But it is simply a good illustration of >>>>>just how many special cases there are in endgames that computers have to >>>>>handle... >>>>> >>>>>Be interesting to see if any program currently has this evaluation knowledge >>>>>to call this a draw, statically. It is doable. >>>> >>>> >>>>Old version: >>>> >>>>material evaluation................. -2.00 >>>>development......................... 0.00 >>>>pawn evaluation..................... -0.24 >>>>passed pawn evaluation.............. 0.00 >>>>passed pawn race evaluation......... 0.00 >>>>king safety evaluation.............. 0.00 >>>>interactive piece evaluation........ -0.92 >>>>total evaluation.................... -3.16 >>>> >>>> >>>>new version: >>>> >>>>material evaluation................. -2.00 >>>>development......................... 0.00 >>>>pawn evaluation..................... -0.24 >>>>passed pawn evaluation.............. 0.00 >>>>passed pawn race evaluation......... 0.00 >>>>king safety evaluation.............. 0.00 >>>>interactive piece evaluation........ 2.24 >>>>total evaluation.................... 0.00 >>>> >>>>Fix was straightforward to do without hurting performance... Just had to >>>>"explain" to it that the right bishop is no good if the opponent has one >>>>too... >>> >>>You need to be carefule here - this position is won for black. >>>[D]8/8/p2b1B2/8/8/pk6/8/1K6 w - - 0 1 >> >> >>I had this one covered already. King must have access to a2 in this position >>for it to kick in the 'drawn' scoring code... Otherwise it was too messy as >>the pawn and king are too close... > >It also was the march of the black king wich did do the trick in the game. > >But it was not a bug in Crafty because schredder5.32 made the same mistake. > >Regards Marc Not a "bug" in the usual context. But a "bug" in a chess context because it was a missing bit of knowledge that screwed this up. This is one of many such kinds of positions where search simply can't do the job, ever, and specific special-case knowledge is required to avoid embarassing moves.
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.