Author: Richard A. Fowell (fowell@netcom.com)
Date: 07:58:59 03/06/00
Go up one level in this thread
On March 06, 2000 at 01:37:23, Bruce Moreland wrote: >On March 05, 2000 at 23:16:54, Richard A. Fowell (fowell@netcom.com) wrote: > >>On March 05, 2000 at 18:25:04, Djordje Vidanovic wrote: >> >><snip> >>>Ernst, hello again. I used my PII-400 under Win 2000 and here's the Shredder 3 >>>analysis (Shredder sees the light after about 12 seconds, that is the march >>>towards the enemy King): >><snip> >> >>Will Bryant's freeware chess program, "Screamer" for the Macintosh, >>( http://pweb.netcom.com/~wbryant/screamer.html ) >>finds the king march in 4 seconds on my 180 MHz Macintosh, and >>mates Black in 13 moves in selfplay at 5 seconds/move >>(that would be 2.25 sec/move on a 400 MHz machine). >> >>Note White's moves 1-6: White Queen covers Black's Queening square, >>then White's King charges straight for Black's King. >> >>I've included here both Screamer's self-play game, and its analysis >>log when given a long time to examine the position. >>Screamer sticks with the king march plan from 4 seconds on. > >The scores returned indicate that it doesn't know what it is accomplishing. >This is the kind of problem where you'd like the program to understand that it >is winning, from the root. > Quite true - the eval makes it clear Screamer doesn't see the end. And I suspect that the eval terms that cause it to win here (centralization terms, perhaps?) would lead it astray in other situations. However, humans pride themselves on winning based on strategic principles rather than direct search, so in that sense, Screamer's win here is "human-like" in some sense. Richard A. Fowell (fowell@netcom.com)
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.