Author: Robert Hyatt
Date: 15:35:18 05/07/01
Go up one level in this thread
On May 07, 2001 at 16:53:12, Uri Blass wrote: >On May 07, 2001 at 16:27:19, Robert Hyatt wrote: > >>On May 07, 2001 at 15:43:50, Uri Blass wrote: >> >>>On May 07, 2001 at 14:43:05, Robert Hyatt wrote: >>> >>>>On May 07, 2001 at 12:30:30, Vincent Diepeveen wrote: >>>> >>>>>> >>>>>>No, he didn't "prove" anything at all. You can't take output from a program, >>>>>>output that is incomplete due to the way they produce their PV, and then try to >>>>>>draw conclusions from it. Any more than you can take evaluations and draw >>>>>>conclusions from them without any idea of what the evaluation looks like >>>>>>internally... >>>>> >>>>>It's hard to argue with scores produced Bob!\ >>>> >>>>Why is that? Crafty vs Diep. Crafty says +2.5, Diep says -.2, Crafty >>>>wins. Big score difference. You do things (queen is way better than 2 >>>>rooks, for example) that I consider unsound. >>> >>>This is a different case because there is a clear explanation for the difference >>>in evaluation. >>> >>>I do not find a good explanation for the difference in the evaluation between >>>deeper blue and the top programs and I see it as an evidence that deeper blue is >>>simply worse. >> >>I try not to jump to conclusions when I don't have any evidence. Often as not, >>I am wrong... >> >> >> >> >>> >>>I do not believe that we can learn nothing from the logfiles of deeper blue. >>>I need to see good moves of Deeper blue when all the top programs need hours to >>>find in order to be convinced that deeper blue was stronger than the top >>>programs of today. >> >>I don't say we can learn nothing. I do say that you can't take their scores, >>and their partial PV, and try to compare them to today's program outputs. >>Because DB's PVs were always shorter than what was actually searched. Trying to >>speculate on what they saw, to explain a score they displayed, is futile... > >I am looking only at the first plies of the pv. >I do not suggest using the length of the pv. > >Uri If you only look at the first few plies, how can you tell whether a +2 score is material gain or positional gain? That is why I think it is very difficult to learn much about their program without having it around. IE it would be nice to have it on my desk, where I can give it positions and let it search to various depths to see the scores for different moves as I fiddle with things... But all you have are the logs with incomplete PVs and scores that come from deep inside the machine. Trying to figure out whether they are missing something or seeing more than another program sees is _very_ difficult to do...
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.