Author: Uri Blass
Date: 06:59:40 10/10/00
Go up one level in this thread
On October 10, 2000 at 08:13:01, Graham Laight wrote: >On October 10, 2000 at 07:31:37, Uri Blass wrote: > >>On October 10, 2000 at 07:05:45, Graham Laight wrote: >> >>>It seems to me that PCs' results against GMs are tapering off into a flat line. >>>The current style of program may have come as far as they can go. >>> >>>The battle to generate the highest NPS score is no longer improving the >>>computers' performance against humans. Even Deep Junior running on a quad >>>processor is only able to score 4.5/9 against the top players. >> >>Only??? >> >>4.5/9 is a wonderful result. > >Of course it's a wonderful result! > >>No longer improving the computers performance??? >> >>4.5/9 against players with average rating of 2700 is the best result of >>computers against humans(if I do not include the result of Deeper blue). >> >>It is even better than the result of deep blue(1996) against kasparov. > >I agree with everthing above. However, on a quad PC, with full 5 piece >tablebases, it should have been challenging for the title, not proud to be mid >table. Single processor PCs may have been able to do just as well if they had >been given the opportunity. Genius 3 beat GK and others (at G/25) about 9 years >ago. It was game/25 and kasparov thought that he was playing genius2. Other programs got worse results against humans and I did not see a reason to expect a very big improvement. 100 elo improvement relative to single processor is very good and Deep Junior got more than it because programs never got performance of 2600 before. > >While 4.5/9 is a brilliant performance against average 2700 elo, it was >disappointing in terms of what one would have hoped for with this setup. I expected it to do less than 50% with this setup so I do not think that 50% is disappointing. > >>> >>>With dozens of programmers competing to make the "final push" to get programs >>>ahead of humans, to impartial observers it looks like the harder they push, the >>>more the bandwagon gets stuck in the mud. >>> >>>Programmers also have to remove knowledge from their eval fns to score higher >>>against their computer opponents. >> >>This is your opinion. >>This is not the programmers opinion. >> >>I see that Programmers add knowledge to their programs in order to have better >>score against computers. > >Ed Schroder has told us that Rebel Century can have a knowledge setting of >several hundred. However, to get best results against other computers, the >optimum setting is 25. This is the case for Rebel but it is not always the case with other programs. I think that one of the problems with Rebel is that it changes its mind more often with bigger chess knowledge and the result is that it cannot get big depth(the problem is not only nodes per second). It is interesting to know if Century with knowledge=500 is better than century with knowledge=25 assuming the same number of nodes(I am not sure about it) >>GambitTiger has knowledge about king safety and I can see it winning computers >>by sacrifices that other programs do not understand. > >That's GOOD news! > >Has GambitTiger won any computer v computer competitions? Yes It won the french championship with 8 out of 9. Gambittiger is not slower in nps and the programmer tries to add knowledge without doing the program slower. > >>> >>>Looks like a doubling of NPS no longer provides an extra 50 Elo rating against >>>humans - nothing even close, in fact. >> >>I am not sure about the nothing even close. > >>>In other words, shooting up, plateauing for a while, then shooting up again - >>>and so on. It's possible that, because chess programmers vary the amount of >>>expertise between 20 and (say) 500 distinct pieces of knowledge, they've found a >>>plateau (probably the 2nd one), and, angry about being beaten by someone with >>>less knowledge but higher NPS, have refused to go down the knowledge route >>>seriously. Also, from many years of reading postings in this group, it is >>>apparent that NPS, and techniques to raise it, is where the focus lies with this >>>particular group of people. > >If this thread continues, I think it will be important to have the diagrams - so >here they are again: > >I think we mostly assume that the return on both knowledge and speed look like >the picture below: > > | > | > | ** > | ************************** > | ** > ELO | * > | * > | * > | * > | * > | * > | * > |* > |* > |* > |--------------------------------------------------------------- > Either Speed or Knowledge > > >But what if, in reality, one or both of them actually looked like this? > > | * > | * > | * > | * > | ** > | * > | * > | * > | * > | * > | ** > | * > | * > | * > | * > | * > | * > | ** > | * > | * > | * > | * > | * > ELO | * > | * > | ** > | * > | * > |* > |* > |--------------------------------------------------------------- > Either Speed or Knowledge > > >>I disagree. >>I know cases when the new version of chess programs have smaller nps. >> >>One example:Fritz6 is alower in nps than Fritz5.32 >> >>Uri <snipped> >But can you put your hand on the bible and swear that this is because of >significant extra knowledge, or might some of it be because franz has discovered >that at this speed, one must reduce the amount of root processing (for example)? I do not know. I am not the programmer of Fritz but I believe that it has more knowledge. 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.