Author: Enrique Irazoqui
Date: 06:08:12 02/01/00
Go up one level in this thread
On February 01, 2000 at 08:07:18, Albert Silver wrote: >On February 01, 2000 at 00:58:37, Christophe Theron wrote: > >>On January 31, 2000 at 09:38:21, Enrique Irazoqui wrote: >> >>>On January 31, 2000 at 09:30:27, Amir Ban wrote: >>> >>>>On January 31, 2000 at 07:48:02, Albert Silver wrote: >>>> >>>>>On January 30, 2000 at 13:36:38, Hans Christian Lykke wrote: >>>>> >>>>>>- Games Won Oppo >>>>>> ------ --- --- ----- --- ---- >>>>>> 1 Junior 6.0 128MB K6-2 450 MHz 2706 38 -35 403 70% 2556 >>>>>> 2 Chess Tiger 12.0 DOS 128MB K6-2 450 MHz 2682 37 -35 414 67% 2555 >>>>>> 3 Nimzo 7.32 128MB K6-2 450 MHz 2661 36 -34 420 68% 2530 >>>>>> 4 Fritz 5.32 128MB K6-2 450 MHz 2659 34 -32 474 65% 2548 >>>>>> 5 Junior 5.0 128MB K6-2 450 MHz 2629 35 -33 438 63% 2532 >>>>>> 6 Nimzo 99 128MB K6-2 450 MHz 2627 44 -42 278 64% 2526 >>>>>> 7 Hiarcs 7.32 128MB K6-2 450 MHz 2624 40 -38 347 66% 2505 >>>>>> 8 Chessmaster 6000 64MB P200 MMX 2574 61 -53 184 76% 2378 >>>>> >>>>>Wow. Sorry if I forgot to convey my own congratulations to Amir and Shay. Like >>>>>everyone else who doesn't yet have the program (notice the "yet"), I am >>>>>extremely curious to play this newest of the wunderkinds. Afterall, it is the >>>>>first to clearly sidestep the results of certain secret tests of certain >>>>>secretive Spanish testers, showing a true evolutive step towards that elusive >>>>>superiority of knowledge over tactics, allowing it to 'see' beyond its tactical >>>>>boundaries. And not content with this you even released a patch improving it! My >>>>>hat is off. Congratulations once more! >>>>> >>>>> Albert Silver >>>> >>>>Thanks very much. And, yes, I was going to about the "secret test suite", but >>>>you beat me to it. >>> >>>So the test goes to its original purpose. Still, all programs perform basically >>>the same in the test and in the SSDF list, as Bertil and Christophe can confirm, >>>and this includes Junior 5. The exception is Junior 6. Why! :( >>> >>>Enrique >> >> >>No offense intended to the new #1, but your test does not take book quality and >>book learning quality into account, Enrique. >> >>I have still faith in your test. It is too good to be damaged by one exception, >>and the exception can be maybe explained by other factors. >> >> Christophe > >Of course you could be right, but that means it would also be the first time a >program's books (and learning) is so good that it defies Enrique's test suite. >Afterall, the other programs all had excellent books, and apart from Tiger >_some_ book learning. If it were just some new book learning (I doubt it is just >the book) then he still has found something so well conceived, it is clobbering >the competition. I realize finding Tiger (for DOS) fell victim to this a number >of times must be quite annoying, but I don't believe this is the biggest issue >here. Forget the results for a minute and look at the games in Enrique's >tournament. I haven't seen it constantly better in the opening, killing the >opposition as a result of it. It has been taking equal positions (sometimes even >worse) and slowly outplaying its opponents. No big power tactics, just better >moves, and Enrique's suite tests tactics. I agree with you, Albert. Looking at the games it is clear that the book doesn't explain the difference test-SSDF. What amounts to be unique to Junior 6, but not to prior versions of Junior, is that it is the only program that deviates from my test. Junior 5 doesn't, since the difference between its performance in my test and in the SSDF list is within 2 Elo points, as shown in the list that I posted here and emailed to Bertil and Christophe *before* the SSDF included programs like Tiger and Nimzo 7.32. All other programs in that list fit well in the correlation test-SSDF. Enrique > I could be completely wrong of course, >but if I'm not, then we are looking at the first program (according to Enrique's >suite) whose knowledge actually outweighs (not compensates) its tactical >deficiencies. I know you dislike the whole search/knowledge comparisons, >claiming they are the same, but it is simply wrong unless the search can reach >deep into the endgame or avoid positional mistakes because of its depth. >Sometimes, it CAN make up for it, but some mistakes may only become a real >problem MUCH much later, far beyond the scope of the search, and can ONLY be >avoided through knowledge. Perhaps when computers can search so deep they can >see this, knowledge can be removed, but for the moment, this is the reality I >perceive. I won a game in the endgame against Fritz 5.32 which should have been >a dead draw, but it allowed me to break it's pawn structure, then tried to hold >on to certain material instead of seeking greater activity (rook ending), and >was finally killed with an eventual king infiltration. Knowledge would _easily_ >have saved it. The point is only to illustrate how important knowledge can >really be. I played the same ending out of curiousity against Hiarcs BTW, and it >_easily_ held the draw. Maybe you're right. Maybe it IS the book and the >learning, but maybe it's something else. > > Albert Silver
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.