Author: Robert Hyatt
Date: 05:55:03 04/01/98
Go up one level in this thread
On April 01, 1998 at 05:57:49, Mark Young wrote: >On March 31, 1998 at 14:38:19, Bruce Moreland wrote: > >> >>On March 31, 1998 at 09:48:03, Mark Young wrote: >> >>>Got my 4 games in with Ferret (runs on a alpha 533). Fritz 5 scored 1 >>>win 1 loss 2 draws. Here are the new results. >>> >>>After 23 games with an avg. player rating of 2573.739 Fritz 5 has scored >>>74% for a rating of 2754.927. >>> >>>Here is the game Ferret won playing Fritz 5. A well played game by the >>>strongest program on ICC. >> >>I don't know that it is the strongest program, but it's been able to >>keep the highest rating for whatever reason. >> >>And yes, this is the Alpha again. >> >>I didn't see the start of this game, but when I tuned in, my program was >>down a pawn but had some counterplay. >> >>I don't know why Fritz sacrificed the exchange, I think it may have been >>a matter of a few centipawns of evaluation. >> >>There was another game that featured a K+P ending that I thought I was >>better in, but when I came back to check the score was like -8. I >>haven't looked at that, I'm not sure I want to :-) >> >>bruce >-------------------------------------------------------------- > >The king pawn endings are like tossing a coin in computer chess. >Ferret's king just lost the race to other side of the board because it >had to capture a pawn first before moving to the other side. Computers >tend to play king pawn endings poorly because of the very long look >ahead needed. Im not sure there is a way for you programmers to give the >computers a way to understand this type of ending better, other then >just ply depth. > > >If there is a stronger program hardware combo on ICC I have not played >it. >When Scrappy gets up to full strenght it may surpass Ferret. But I >understand it is very hard to get an optimal algorithm for parallel >processing. So we will just have to wait and see what Bob can do, but Im >sure with his skill in this area he will get crafty much stronger. > > Mark Young this is fixable. I have specific code in Crafty to handle the "distant passed pawn" problem. if you reach an endgame playing Crafty, and you have (white) pawns on a2,b3,c4, and king at b1, and Crafty has pawns (black) at c5,b6 and h7, it "knows" it is winning that game because your king has to stop the passed h-pawn, giving it time to eat your pawns on the queen-side... I don't yet recognize a "majority" although it is a special case of this same idea...
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.