Author: Robert Hyatt
Date: 08:43:51 11/12/97
Go up one level in this thread
On November 12, 1997 at 00:23:28, Robert Sherman wrote: >On November 11, 1997 at 09:29:35, Robert Hyatt wrote: > >>On November 10, 1997 at 22:01:54, Robert Sherman wrote: >>>> >>>I am impressed that Crafty steers itself out of trouble. But a >>>seven-point swing for a computer program based on positional safety of >>>the king is incredible. It certainly makes that Deep Blue Be4 move >>>totally normal because computer programs were only a point or two >>>different. Your score is 7-8 points different from all the commercial >>>programs that I have seen. >> >>I am re-running this and will post the results soon. A couple of >>points, >>however... >> >>1. I do *not* have king-safety terms that cause a +/- 7 pawn swing. :) >> >>2. I might have posted bogus analysis. I get hung up with position >>learning >>about once per month. While annotating, position learning is disabled, >>but I >>played around with this game before doing the annotate, and it is highly >>likely >>that Crafty "learned" something along the way that screwed the scores up >>at >>this point in the game. >> >>I will post a new annotated game in a few mins to see how it looks after >>the >>position.bin file has been deleted from my notebook... > >I look forward to seeing your new analysis. Regarding your first point, >I thought so, I have never looked at your code thoroughly, but what I >saw of it didn't seem to show any kingside safety heuristics that would >produce such a result. So, I guess it will have to be related to your >learning function. >Still I am impressed with the "amateur" Crafty sidestepping an >aggressive but unsound attack. I think the proper candidate moves in >this position are f5, f 6 and d6, but o-o survives with the computer's >nerves of steel. I am sure your analysis will show that. > >Robert Sherman I have not yet been able to re-annotate the entire game. But O-O is now producing a reasonable score, and if it had to play O-O, it would follow it with g6 as you suggested. And the score is now fairly close to the other moves, rather than being -8 worse. :) The position learning is a problem. I generally have everything on, as the default. However, I'm thinking about making this learning option "off" by default as I get plenty of questions about the results this can cause. If you play down the O-O line a ways with position learning on, it can remember a bad position, and then when you back up and try O-O, the result of this bad position can filter into the evaluation and make analysis *very* confusing. It works great for those players that want to learn a particular opening, and discover that Crafty isn't very well booked for this particular variation. It will still "learn" what is going on and continually find new and better ways to defend, until it "solves" it or you continue to find different ways to continue...
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.