Author: Uri Blass
Date: 01:23:26 02/14/02
Go up one level in this thread
On February 14, 2002 at 02:46:36, Slater Wold wrote: >On February 14, 2002 at 01:36:08, Uri Blass wrote: > >>On February 14, 2002 at 00:20:56, Slater Wold wrote: >> >>>On February 13, 2002 at 20:34:27, Dann Corbit wrote: >>> >>>>On February 13, 2002 at 20:29:42, Slater Wold wrote: >>>> >>>>>Ok, so I have decided to run the NPS challenge all over again. And this time, >>>>>there will be a winner. >>>>> >>>>>HOWEVER, there will be a few differences. No one seems to have a version 11.x >>>>>Crafty, and I don't have a computer to run the old Rebels. So this is not by >>>>>choice, rather by must. >>>> >>>>ftp://cap.connx.com/pub/chess-engines/new-approach/Ancient-Crafty/ >>>> >>>>It's very balky under Winboard. I wouldn't recommend it anyway. >>> >>>That was going to be my next question, if anyone actually had it. >>> >>>>>Here it is: >>>>> >>>>>Century 4 vs Crafty 18.13 >>>>> >>>>>20 games of 40/120. Just to get a baseline. >>>>> >>>>>20 games of Century 4 @ 40/2 hours vs Crafty 18.13 @ 40/200 hours. (Exactly >>>>>100x the amount of time Century will have.) >>>>> >>>>> >>>>>What this match will prove: Not a damn thing. >>>>> >>>>>But you have to admit, if Crafty loses horribly in the first match, but does >>>>>outstanding in the second match, wouldn't it raise your brow? Would mine. >>>>> >>>>>Century will get approx. 3 minutes per move, Crafty will get approx. 200 minutes >>>>>per move. That is *quite* an advantage. >>>>> >>>>>I will post the games after each one is played. With a current result. PGN's >>>>>and logs will be available by request. >>>>> >>>>>Ok Ed, up to you now. All I need is Century 4. :) >>>>>(I have Century 3 by the way.) >>>> >>>>I think we will see a butt-blasting in Crafty's favor. Crafty's search is a lot >>>>smarter now. A 100 minute per move straight-up match would be more interesting. >>> >>>'Ya think so? I would have to say that Rebels search would have to be improved >>>SOMEWHAT. Perhaps no where near Crafty's improvement over the years, however, >>>it has been approved. >>> >>>If it's a "butt-blasting", then I will prove what everyone has known. Faster >>>hardware = better chess playing. >>> >>>>Did you see Ed's chess in 2010 match (something like that). >>>>Crafty was doing *very* well in it. So extending the time for Crafty and >>>>contracting the time for the opponent will just mean a one-sided whitewash. >>> >>>No I didn't. Perhaps I will look into it. And maybe........just maybe....after >>>this match, I will repeat the same with Crafty 11.x and C4. Or Shredder 6 vs >>>Little Goliath. >> >> >>I see no reason to use shredder6. >> >>I believe that Fritz7 is better. >> >>I also do not see a reason to use little goliath because it has not a simple >>evaluation. >> >>I believe that the top programs main advantage against Crafty is better search >>rules. >> >>If the target is to test evaluation against search then I think that the best >>solution is simply to test Crafty against modified Crafty with small evaluation >>and the same search rules. >> >>The source code of Crafty is free so I guess that it is easy to change the >>evaluation in order to have simple evaluation (for example only piece square >>table). >> >>Maybe I guess wrong because I did not learn the source code of Crafty so I do >>not know if there are some assumptions about the evaluation in the search rules. >> >> >>Uri > >The original point of the challenge was to prove that a program with a simple >search on SUPER HW, could indeed beat a complex program on much slower hardware. > (Ah hum, Deep Blue vs Rebel was the case and point.) > >I agree that Crafty is no Deep Blue. And thinking for 200 minutes between moves >will not make it close to Deep Blue. > >Back then, Crafty 11.x has a very simple search. Today, it is not so simple. > >My main thinking here is: > >Rebel will more than likely win 75% of its games against Crafty in the 40/120. >Now if Crafty can beat Rebel 75% of the time in the 40/12000, what does that >show? That Crafty is "rebel strength" when given a 100x time odds? Or that >simply HW = chess program performance? Well, I think all that will be >determined in the moves in makes in the time odds games. > >Little Goliath vs Shredder was something that GCP suggested. LG is extremly >fast, and from what I understand, not overly complex. The programmer said that he used assembler to optimize part of the move generator and I understood from posts of the programmer that the program is complex. real simple programs like tscp have no chance even if they get 200 hours/40 moves against 2 hours/40 moves of the opponent and tscp is not the worst program in the world. In the case of tscp it is also because of a design decision of the programmer so you can expect the program to crush or to be slower if you try to play it at 200 hours/40 move time control but even if you correct the problem tscp still has no chance against the top programs if you decide 2 hours/40 moves against 200 hours/40 moves(tscp has worse branching factor and I know that it is in the similiar level like palm tiger at fast time control) The hardware difference between palm tiger and modern hardware is big(it seems that modern hardware is close to be 100 times faster than the palm) I believe that even if you take gerbil that is a simple program that is better than tscp then it is going to lose in a match when it gets 200 hours/40 moves against 2 hours/40 moves of top programs. Shredder on the other >hand is as slow as they come, and again, from what I understand, very complex. I do not see a reason to assume that shredder is more complex than Fritz. 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.