Author: Robert Hyatt
Date: 20:18:09 05/26/99
Go up one level in this thread
On May 26, 1999 at 23:14:59, Robert Hyatt wrote: >On May 26, 1999 at 20:22:28, Paul Richards wrote: > >>On May 26, 1999 at 19:32:02, José de Jesús García Ruvalcaba wrote: >> >> >>> The main problem is that no program will see the repetition, as white >can easily make 50 moves without allowing a third repetition. The program >>>would need to detect the 50-moves draw with the search, and that would not >>>be enough as white can make five pawn advances, resetting the count. >>> Then the program needs to search 50x5x2=500 plies to detect the draw, >and that is quite far from what current programs reach on current hardware. >> >>Crafty completely ignores gxh6?? as an option, and if forced to make the move, >>realizes the position is drawn at 16 ply (this is within the Fritz GUI, >>no tablebase access enabled. So what is the code that allows it to avoid >>gxh6?? immediately, and how does it determine the draw after making the move >>gxh6? This would be rather easy for Dr. Hyatt to explain, hence the use of >>the word Crafty in the subject to draw his attention. :)) In any case most >>programs do see the draw, and without resorting to tablebases. > >To see a draw, in this kind of case, you need two circumstances: The side >that is 'seeing' the draw hopefully thinks it is better. The other side, >therefore, thinks it is losing. > >Given those two, the side trying to win discovers that if it doesn't play the >move leading to a draw it loses outright. But by playing the move that does >lead to a possible draw, the opponent can't find any way to win, and since that >side thinks it is losing, it heads to the draw a.s.a.p. Not knowing that it >can also play a much longer draw if it wants. > >Cray Blitz would handle this far differently, but it would produce the same >result. If the other side (in this position) was well ahead, then the other >would not be able to force a draw it could see, because the winning side would >try like mad to avoid the draw and push it off so far into the future that it >would not be seen. In essence this is a position where the drawing move is the >only move that doesn't lose, and by playing this the other side becomes >convinced that taking the draw is the right thing to do. > >Almost like _both_ sides think they are losing, but not quite. :) also I didn't look at the position, but there are many positions that Crafty knows about... ie bishop + rookpawn of the wrong color. So that it can evaluate something as drawn even though it is way up in material. Perhaps Fritz doesn't know about this. It is amazing how many programs don't know about not playing Bxa7 if black can play b6 and trap the bishop. But I watched Crafty spring this trap on two well-known programs yesterday and today on ICC. B+RP is just another important piece of knowledge that will save a game here and there at nominal execution time cost.
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.