Author: martin fierz
Date: 17:44:38 04/12/02
Go up one level in this thread
On April 12, 2002 at 17:05:45, Mark Young wrote: >On April 12, 2002 at 14:28:13, martin fierz wrote: > >>>On April 12, 2002 at 03:43:16, Mark Young wrote: >>> >>>Sorry,Fritz 7 could be stronger,NPS means nothing between 2 programs >> >>NPS does not mean everything, but it does not mean nothing either. > >I did not say that it means nothing, I said NPS means nothing between 2 programs >that are not the same. > > > typically, >>NPS is a "design choice" of the programmer - smart but slow or less smart but >>faster. > >This is clearly wrong here, are you selling us that you can tell how smart a >program is by NPS? Your understanding is limited, if this is your bias. my understanding is quite all right, thank you :-) NPS is very much a function of how smart a program is. of course, if you write a program in basic, and get low NPS, this does not mean that your program is smart! i never said that. if you compare well-optimized engines (take some of the top commercials, fritz, junior, shredder, hiarcs, whatever), THEN NPS is a measure of how much work the program does on every node. in my checkers program, for example, i can turn on or off several features like enhanced transposition cutoffs, evaluation-based forward pruning and arbitrary parts of my move ordering. i can also add or remove parts of the evaluation function. everything i turn off makes my program faster in NPS, but it plays worse. the difference between smart and dumb is a factor 2 in NPS for my program. this is the design choice i was talking about, and it is the reason why people here always say "NPS means nothing". if you are talking about this kind of difference in NPS, it *does* mean nothing. talking about a factor 50, without the penalty of being dumb in return is another matter entirely. >How strong was Deep Blue 1997? How strong are todays micros? If you look at the >facts based on games played....It is clear your argument does not hold water. >What is Clear based on the data. Todays top micros are playing better then 2700 >elo chess. And playing very close if not better then DEEP BLUE of 1997. facts based on what? like rebel-van wely 2-2 or huebner-fritz 3-3? i don't think that these results are "very close" to defeating kasparov by 3.5-2.5! look at what kasparov does to these 2 guys: searching the big2000 database for kasparov - van wely gives me 4 games at standard time control - 4-0 for kasparov. in a match against huebner in '85 kasparov won 4.5-1.5. in matches at STC, where humans prepare and learn, i have yet to see "better than 2700"... i know no other serious results than the 2 above. >That why >NPS means nothing....Based on the facts and your logic, Todays top micros must >be pure Genius next to Deep Blue as Deep Blue needed 100X to 1000X the NPS that >the top micros needed to play about as well. it looks like the null move and other forward-pruning techniques have helped a lot to make programs better. i agree with you there - this is something DB didnt have at the time AFAIK. but keep in mind that these are basically unsound techniques, mostly they work and improve your program, but sometimes it's the opposite... there have been many posts here with positions in which programs could fail spectacularly thanks to forward pruning techniques. DB didn't have these issues. aloha martin
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.