Author: Vincent Diepeveen
Date: 05:09:59 10/11/02
Go up one level in this thread
On October 11, 2002 at 08:03:04, Uri Blass wrote: >On October 11, 2002 at 07:51:42, Vincent Diepeveen wrote: > >>On October 11, 2002 at 07:43:40, Uri Blass wrote: >> >>But he's an old idiot. He's still talking about computerchess, >>forgetting how bigtime the weakest chain has gone up last few >>years. In 1997 games were decided by programs blundering away >>material. All programs were very weak in endgame by then too, >>any random rook endgame i could win from any program in 1997. >> >>How things have changed there... >> >>Even theoretical you can proof that without nullmove and >>without 'junior' type of forward pruning you can't get >>18 ply fullwidth at all. >> >>Knuth in fact proved a lot about that already. It's so >>easy. Only because this guy has 'professor' before his >>name doesn't mean that he can do something that theoretical >>is impossible. >> >>Apart from that the statements from the deep blue team >>are very clear in 1999 in the IEEE advances where they >>show a 12 ply search. The 12.2 average depth in their >>artificial intelligence thesis. >> >>Then last but not least. No one got 12 ply in those days. >> >>I only remember fritz3 which by very dubious means got like 11 >>ply. Basically some preprocessors did get 11 to 12 ply thereby >>forward pruning last few plies a lot. >> >>None of them had things like simple mobility in the leaf eval >>even. >> >>Now deep blue got in the past when it was deep thought with >>500k nodes a second after 3 minutes sometimes a search depth >>of 8 ply. With 126MLN, from which like 95% was wasted to >>parallel search, they got 12.2 ply. > >This is another question to ask Hsu. > >What is the effective number of nodes of deeper blue? >Is it 126M/s nodes or does he believe that with one processor and 126M/s nodes >he could do better. > >Again people disagree here > >Uri you skip one important point. Because of a simplistic evaluation it was able to get 12.2 ply. If you use a more complex evaluation then you do fullwidth not get 12.2 ply at all, but more like 10.5 ply. I have hard proof from DIEP here. I remember how i easily searched real deep with a 1999 version in endgame. I searched 20 ply with nullmove there in endgames at bob's quad. TWENTY ply. However it was because in those times diep's endgame wasn't worked at at all. I just worked at opening and middlegame knowledge. Never endgame endgame evaluation was like endgame(int STM) { return materialscore + kingincenter[kingpositions]; } I simply never cared until 1999 for a program being strong in endgame. Therefore i searched 20 ply. We see deep blue do stupid moves, it never doubts in most cases about the move it plans to play. Otherwise with all those extensions they would never get further than 11 ply at all. So the comparision is already not very fair at all. If deep blue gets upgraded to 2002 standards, its evaluation especially, then it will be from search depth viewpoint with fullwidth search, a complete joke. Of course it gets completely annihilated when appearing in 1997 standards. So if Hsu upgrades his chip to a single cpu chip with a new and better evaluation (it's of course questionable whether he is capable of manaqing that) then it will not search deeper than deep blue in 1997 of course, unless he adds nullmove and hashtables. Best regards, Vincent
This page took 0.01 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.