Author: Uri Blass
Date: 12:40:22 11/18/01
Go up one level in this thread
On November 18, 2001 at 12:43:48, Jesper Antonsson wrote: >On November 18, 2001 at 08:49:06, Uri Blass wrote: >>On November 18, 2001 at 08:13:48, Jesper Antonsson wrote: >>>On November 18, 2001 at 05:06:24, Otello Gnaramori wrote: > >>>Well, you don't show anything there to back up the claims that their algorithms >>>are outdated. No new revolutionary algorithms have been presented that would >>>make DBs outdated, >> >>It is known that DB did not use the null move pruning. >>The null move pruning was known in 1997 but they did not use it. > >Yes, I know. And if they didn't use it, they probably had good reasons to. So, >no new revolutionary algorithms since then, as I said. Their reason may be believing in the wrong assumption. At their time the best program(Genius) did not use null move pruning. > >> >> and how complex/good their eval is we don't know. Since they >>>could make complex eval for free in hardware, it might just be much better than >>>todays micros. >> >>I do not believe that it was better than today micro. >>I analyzed the games andthe fact that deeper blue could not see Qe3 in the pv of >>game 1 suggest that their search rules and their evaluation was inferior. > >Could you explain why one single situation would make you draw such far-reaching >conclusions? I did not say that I am sure about it but the fact that I could not find impressive tactical move that they found and the fact that they missed the drawing move in game 2(other programs also do not see that there is a draw but they at least can see the move for positional reasons) suggest that they were inferior. > >>I saw no evidence for a move that top programs of today need a long time to find >>and other people also analyzed the games and found no evidence for it. >> >>It seems that every move in the games can be played after some minutes by one >>of the top programs(not always the same program). > >Exactly, not the same programs. So I can prove that Fritz has no better eval >than a random move generator, since some of my ten thousand random move >generators will play the same move as Fritz? If what you describe here is your >analysis, then I think it's too shallow and fundamentally flawed in it's >methodology. I did not say that it is a proof but it supports my impression. The top programs are not random move generators and if something is really clearly better than them then I expect at least one move that no top program can find because it is too deep for top programs. People can claim that there were not enough games for getting impressive move from Deeper blue so I agree it is not a proof. > >To get something useful, I think one should compare *one* engine at a time, see >what moves differs, and thoroughly analyse those moves to see 1) if the micros >choice is better, 2) if it sees it for the right reasons or just by "luck". The >results would say something about which machine is better, but what you describe >does not. I agree that it can be productive to do it but the problem is to decide which move is better and people may disagree about it. > >>>Perhaps a real expert may be able to say something about this if he was to >>>scrutinize the DB moves in the last match and compare them to the moves Fritz >>>would make if given a lot of extra time. But it would be hard to be fair there, >>>since "nodes" aren't equal. >>> >>>Regards, >>>Jesper >> >>My impression based on analysis that Deeper blue was only 2-3 times faster than >>Deep Fritz on p800 in some cases because deep Fritz was 2-3 times slower in >>finding similiar pv to Deeper blue. > >Always? Remember, since they didn't use null-move, they might be slower than >they would be otherwise, average case, but find some nice gems quicker. I admit that in most of the cases even when it wanted to play the same move it did not find similiar pv's but in few cases when it found similiar pv's it was 2-3 times slower. I analyzed some positions for many hours and never found cases when Deep Fritz(p800) was 10 or 100 or 1000 time slower than Deeper blue in finding similiar pv. > >>There are logfiles of the games of the last match deeper blue-kasparov so it is >>possible to check it. > >I would if I had the time, but I haven't right now, so I will have to take your >word for it. You have not convinced me yet that your conclusions and methods are >correct, though. > >Regards, >Jesper I admit that I have no scientific proof and it is only my impression based on the games. I believe that if people analyze Deep Fritz-Heubner match they can find moves of Deep Fritz that no top program can find in few seconds. I have no proof for it because I did not analyze the games of Deep Fritz-Heubner with all the top programs. If I am right then it is going to be another evidence(not proof) that my opinion about deeper blue is correct. If I am wrong then it is going to do my evidence weaker. 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.