Author: Vincent Diepeveen
Date: 00:30:36 11/24/98
Go up one level in this thread
On November 23, 1998 at 11:36:44, blass uri wrote: > >On November 23, 1998 at 09:37:25, Robert Hyatt wrote: > >>On November 22, 1998 at 11:49:54, blass uri wrote: >> >>> >>>I did not ask for all the tree but only the tree up to the point that my >>>programs can see by search of 3 minutes that black has at least 1 pawn >>>advantage. >>> >>>This is clearly less positions >>>because if in the leaves it is -2.xx then Junior can see some moves before the >>>leaves that it is -1.xx >> >> >>ok... rather than 10 million pages, it might only be 1 million pages. How >>would we get those to you? :) >> >>what you are overlooking is the point that junior (and all the other programs) >>look at a fat, shallow tree. IE how do you think a program like Crafty, doing >>300K nodes per second, reaches 12 plies in the middlegame, while Deep Blue, >>doing 1,000 times as many nodes per second, only reaches to 10-11 plies in the >>middlegame? Because *they* are searching 10 times deeper than I am on most >>moves, thru their "singular extensions" (and other extensions). We've already >>seen that in the Deep Blue vs Kasparov game two, Dark Thought and Ferret have >>searched axb5/Qb6 to depth 20 or 21 without seeing anything to cause it to fail >>low, yet we know deep blue did. At 1/2 that depth. So it might take a program >>like junior *fifty* plies to find what is going on there for all I know at >>present. And if I could somehow give you a PV to get you down to the point >>where Junior sees this, it would be so deep, probably, that it would be easy >>to say "but this isn't the best move, white or black should try this instead. >>And we end right back up at square zero. >> >>There are just some things they can see at 250M+ nodes per second that we won't >>ever see... > >we are discussing about deepthought 2 game and not about deeper blue >deepthought could not calculate 250M+ nodes per second > >> >> >> >>> >>>>I don't have their "output" for this move. As I said before, we sat at the >>>>same table playing this game at the 88 ACM event (I think). I saw their output, >>>>they saw ours. We both saw them fail high with a score > 2.0, while we were >>>>reasonably happy with our score... until the roof fell in about 10 moves >>>>later... and their eval didn't vary by much for the entire sequence... So I >>>>can't give you their output, since I don't have it (they were using a laptop >>>>to display their stuff). I can only tell you what actually happened in the >>>>game. >>> >>>I believe that cray blitz lost because of a mistake that came after c5(maybe at >>>move 32 because I do not see what is wrong with 32.Bg5) >>>The fact that they have score>2.0 does not prove that they were right in the >>>evaluation. >> >> >>You'll have to believe what you want here. I *know* that a program that doesn't >>do any selective forward pruning and which doesn't use null-move is *not* going >>to make that kind of mistake, except perhaps for some sort of horizon effect on >>the end where they can't actually take the piece due to a mate threat or some- >>thing more serious. But that's not the case in this position... > >There can be a mistake in the evaluation function >They may evaluate -2.xx something that top programs evaluates as -0.xx Exactly. In the sacrafice line, the white kings get to e2. Deep Blue probably evaluates that 1 pawn tougher all together (the king safety), then already without singular extensions you're playing axb5. Difference in evaluation is around 0.90 pawn, so 1 pawn extra for king safety. The heuristic is also easy: just scan around king for pawns: not a single pawn there when king gets in the Qb6 to the square e2. This is not a single proof of singular extensions. All moves it made only are a proof of very few evaluation terms, which are set very high. >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.