Author: Robert Hyatt
Date: 15:00:11 07/28/02
Go up one level in this thread
On July 28, 2002 at 14:29:19, Vincent Diepeveen wrote: >On July 28, 2002 at 12:51:33, Robert Hyatt wrote: > >>On July 28, 2002 at 12:47:36, Robert Hyatt wrote: >> >>>On July 28, 2002 at 02:48:28, Vincent Diepeveen wrote: >>> >>>>On July 27, 2002 at 23:10:10, Robert Hyatt wrote: >>>> >>>>Oh that position, yeah there the Deep Blue search system >>>>which are focussed upon mate threats, they should >>>>work well there. >>>> >>>>Crafty is a bad compare here. >>> >>>How long does it take your program to get +2.5??? >>> >>>Best so far is Hiarcs at several minutes... >>> >>> >>> >>> >>>> >>>>>On July 27, 2002 at 19:57:39, Vincent Diepeveen wrote: >>>>> >>>>>>On July 27, 2002 at 09:02:13, Alvaro Jose Povoa Cardoso wrote: >>>>>> >>>>>>what the hell do you talk about? >>>>> >>>>> >>>>>He is talking about the Nf6+ move crafty played against chessmaster in >>>>>the KKUP tournament a few years ago. Crafty saw a draw score after a 13-14 >>>>>ply search. Hsu sent me some brief output from Deep Blue Junior that showed >>>>>that in under 2 minutes, it saw our score as +2.5... Several moves into the >>>>>game Crafty failed high and finally agreed. DB Jr saw it very quickly while >>>>>everyone else is taking quite a long time compared to their two minutes. Note >>>>>that this was DB Jr model 1, not the 1997 version... >>>> >>>>r1b1r1k1/1q3ppp/ppn5/2bNp3/P4B2/5Q1P/BP3PP1/R2R2K1 w - - 2 19 >>>> >>>>DB jr is just as good as the 97 version, considering their parallel >>>>speedup extrapolation. >>>> >>>>>>I can show you a bunch of bad moves DBII made against kasparov >>>>>>which no other computer makes, but i don't see a single good >>>>>>move DB makes which todays software doesn't find. >>>> >>>>>This was simply an attempt to compare their tactics to ours. At least in >>>> >>>>To that of crafty of course ,which isn't doing many checks in qsearch >>>>nor extending mate threats a lot. nor extending singular stuff. >>>> >>>>in this case the system as they describe how they extend works great >>>>of course. I bet Brutus will find this easily too. >>>> >>>>BTW did Hsu give an output? >> >> >>Yes he sent me output. As I mentioned, the problem is trying to find it >>in my email backups. That was back around 1996 or so, buried an hundreds >>of thousands of emails... Not easy to find. > >that would be in itself weird because the game with Nf6 was played >januari 99 or so. I don't believe so, but I am not sure... I believe this was played in 1997, early. I don't think Alan was still around in 1999. Nor do I think that in 1999 anybody would consider a 486 a "normal" machine. The "benchmark" was a 486/66 machine for 8 hours. I don't think that would be true in 1999... Perhaps you might check your data again, as mine suggests 1997 although the game might have started in late 1996... The tournament took a good while to finish up... Update. I just checked the PGN and it says 1997 so this is probably the right time, although late 1996 rings a bell to me... > >Note that diep sees very quickly a draw in deepblueII-kasparov >for the move Qe3. If you see a draw after Qe3 you have a bug. Particularly if you see it "very quickly"... It is a _long_ way from Qe3 to a forced repetition. > In fact it doesn't play the blunder Kf1?? from >deep blue but plays kh2! instead within tournament time of 40 in 2!! So? Different king safety and endgame king placement scoring. Just because you don't play Kf1 doesn't mean you don't play it for the right reason. Probably serendipity more than anything, as I haven't noticed Diep having a particularly good idea of where the king should go, whether it is being attacked or in the endgame... On general principles Kh2 is horrible, unless it is forced. If you don't see it is forced, it is horrible even though it is the right move. > >Best regards, >Vincent > >> >>>> >>>>because all the things i hear is always like: "he told me". >>>>I want to see outputs :) >>>> >>>>it could be true he just made a few moves and then got the score. >>>>i remember an analysis of it a few years ago. if you give a big patzer >>>>score for king safety you sure can get +2.xx there, otherwise it's a >>>>pretty deep combi to see you win a piece on g6 using a pin of a pawn h4 h5. >> >>He got the +2.5 while searching the given position. His PV started with >>Nf6+ and the score was +2.5 before 120 seconds had gone by. I don't remember >>when it failed high, I just remember the +2.5 took under two minutes. Crafty >>searched to 14 plies on that move in 1996 and had 0.00. It had 0.00 for >>another couple of moves before it started failing hi as Hsu promised... >> >> >> >>>> >>>>>this position, theirs is better. >>>>> >>>>> >>>>> >>>>>> >>>>>>>Sorry to return to this old post by Dr. Robert Hyatt, but I red it a little >>>>>>>late. >>>>>>>I know this is just one position, but I was wondering what conclusions could we >>>>>>>take from this test? >>>>>>>It seamed to me that current programs were a little slow in finding the winning >>>>>>>score (not the move). >>>>>>> >>>>>>>Any comments? >>>>>>> >>>>>>>Best regards, >>>>>>>Alvaro
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.