Author: Sune Larsson
Date: 07:34:15 02/25/01
Go up one level in this thread
On February 25, 2001 at 09:45:48, Uri Blass wrote: >On February 25, 2001 at 08:55:47, Sune Larsson wrote: > >>On February 25, 2001 at 07:55:31, Ferdinand S. Mosca wrote: >> >>>On February 25, 2001 at 07:10:13, Sune Larsson wrote: >>> >>>> >>>> [D]7k/7P/P5K1/8/4B3/8/8/6b1 w - - 0 1 >>>> >>>> >>>> This is a dead draw. Black just moves his bishop on the a7-g1 diagonal. >>>> Correct evaluation of this position is 0.00. Here are the results of >>>> some programs, after a few minutes of thinking on a PIII 800. >>>> It's a falling scale from "best" to "worse". >>>> >>>> 1) Nimzo 7.32 + 1.83 >>>> 2) Nimzo 8 + 1.96 >>>> 3) Hiarcs 7.32 + 2.20 >>>> 4) Chess Tiger + 2.40 >>>> 5) Junior 6 + 2.44 >>>> 6) Fritz 5.32 + 2.75 >>>> 7) Crafty 18.01 + 2.80 >>>> 8) Century 3.0 + 3.23 >>>> 9) Gandalf 4.32g + 3.24 >>>> 10) Phalanx 22 + 3.28 >>>> 11) Fritz 6 + 3.84 >>>> 12) Junior 5 + 3.85 >>>> 13) Deep Fritz + 4.03 >>>> 14) SOS + 4.78 >>>> 15) Gromit 3.1 + 5.40 >>>> >>>> Sune >>> >>> >>>Very impressive comparison, I thought now of doing the same but for a >>>different position, maybe to give a hint of which program packs >>>a particular knowledge. What will be your suggestion on the time to be used? >>> >>>Regards, >>>Dinan >> >> For practical reasons and since this is about knowledge, I used 3 min >> of thinking time per program (PIII 800). I also gave a couple of them much >> more time, just to check, but the evals wouldn't change. If you know you know >> and if you don't you don't...;) > >I disagree. >It is possible to learn to change your evaluation function. Sorry Uri but I think you missed my point. I meant that it doesn't matter if a program thinks 3 or 20 minutes in the above position. Either it has knowledge about it or not. But of course the programmers can change the eval function. In fact, that is one of the points by posting this. Sune > >Humans without previous knowledge may evaluate it as a win for white and only >after analyzing the position >they will understand that it is a draw. > >There is no reason that programmers cannot teach their program to change their >evaluation function as a result of previous search. > >There is also a problem in comparison between numbers because I do not think >that programs means to the same thing by the same number. > >It is easy to teach programs to have more knowledge in the position that you >post by your definition. >The only thing that you need to do is to divide all the evaluations by 2. > >Uri
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.