Author: Robert Hyatt
Date: 08:59:58 11/10/00
Go up one level in this thread
On November 10, 2000 at 11:23:50, Joe Besogn wrote:
>On November 10, 2000 at 10:27:02, Robert Hyatt wrote:
>
>>On November 10, 2000 at 07:46:31, Thorsten Czub wrote:
>>
>>>On November 09, 2000 at 18:33:05, Robert Hyatt wrote:
>>>
>>>>On November 09, 2000 at 15:20:24, Thorsten Czub wrote:
>>>
>>>>>Rf3 } 38...Red8 39.Kh1 { Score: 3,42 depth: 10 Kh1 Nf5 Qg5 Bc6 Bg6 Qf6
>>>>>Qh5
>>>>>Nh6 Rg3 Be8 }
>>>
>>>>OK... Here is a good point for discussion. My evaluation at this point
>>>>is somewhere between -.2 and -1.0, depending on how deep you let it
>>>>search. Here is the position:
>>>
>>>[D]2rr2k1/1p1b2p1/p3p2n/2Bp3B/1P3P1Q/PP5R/1q5P/4R2K b
>>>
>>>Lets see what other programs think about move 39...
>>>
>>>CSTal2.03: d8 88" +1.16 Qg5 b6 Bxb6 Rf8 Bc5 Rf6 Bg6 Qxa3
>>>Shredder4: d11/23 1'23" +1.07 Qe7 e5 Rh4 d4 fxe5 Qc3 ...
>>>Gandalf4: d9 ~ 1' +0.86 Bg6 e5 Rg3 Qd2 Rxe5 Qc1+ Kg2
>>>Hiarcs7.32: d9/27 1'11" +1.84 Qe7 e5 Rg3 exf4 Rg2
>>>Fritz6: d11/34 1'44" +0.72 Qe7 e5 Rh4 d4 Qxe5 Bc6...
>>>Junior6: d16 1'39" +1.06 Qe7
>>>CMaster6000:d3/8 1'46" +0.71 Qg5 Qd2 Qe5 Rc6 Bg6
>>>
>>>>It is black to move. White has the rook and queen doubled on the h file.
>>>>Black has the h-file blocked with the knight on h6. White has no way to
>>>>drive the knight off, and no easy way to capture the knight with some sort
>>>>of trade. IE the h-file is not going to be used for an attack very easily.
>>>
>>>>Do you (or anybody) think that white is really up the equivalent of one
>>>>piece (+3.42) here?
>>>
>>>Why do you believe black is NOT lost here ?!
>>>IMO the position is very clear. black is shortly before execution.
>>>
>>>>Is black blind? Is white over-optimistic?
>>>
>>>I think white is right.
>>>As you can see CSTal + Hiarcs have high evaluations too.
>>>They smell that danger.
>>>Of course Gambit-Tiger sees is much earlier in the tree.
>>>Hiarcs and CSTal are SLOW programs, when i let CSTal and Hiarcs compute
>>>LONGER, CSTal says after 1273s
>>
>>
>>Note that except for CSTal, my eval pretty well agrees with the rest. At
>>the 1+ minute mark, my score is about -1.0, which is right in line with the
>>rest. I notice _nobody_ said +3.5. So I suppose I miss your point here.
>>
>>
>>
>>
>>
>>>
>>>CSTal2.03: d10 1273" +3.33 Qe7 Kh7 Bf7 Qf6 Qxf6 gxf6
>>>Hiarcs7.32: d10/30 34'19" +2.41 Qe7 Kh7
>>>
>>>so the other programs SEE it too, only they see it much later.
>>>
>>>> Or is the truth somewhere in
>>>>between?
>>>
>>>
>>>the truth is IMO that white missed the right way to excute black.
>>>the position is lost for black IMO.
>>
>>
>>
>>So we are back to your old argument about CSTal? IE that it can reach
>>won positions but it can't win them? This chain of reasoning doesn't leave
>>me warm and fuzzy at all.
>>
>>
>>>
>>>Crafty's misevaluation is the problem it came into the position.
>>>and gambit-tigers , as you call it, inflated evaluation is the reason
>>>it played such an attack.
>>>
>>>If we give Tiger more time to consider , it sees that Qe7 is better...
>>>
>>>>My score for move 47 is -.33... it has very slowly climbed over the last
>>>>8 moves. I don't see how this game is a good example of what you call a
>>>>"new paradigm". It looks like the evaluation was inflated, GT slowly
>>>>found that it couldn't hold that position with the somewhat inflated score,
>>>>and the score settled back down closer to what Crafty was saying.
>>>
>>>I want to explain why i believe this game is a good example.
>>>I have seen many many of those games, with cstal and gambit-tiger.
>>>You say: the evaluation was inflated, the score settled back down.
>>>you say so, as if the score is a graph and has to be a line, that should
>>>come very close to what YOU call "reality".
>>>this idea is the old paradigm.
>>>it's like newton and his believe in a universal-time, that is everywhere the
>>>same, and all we have to do is to adjust our watches to this universal-time
>>>that is the same in the whole universe !
>>
>>I certainly don't mind discussing this, but handwaving, shouting, going off-
>>topic don't convince me of anything. IE discussion about reality, old paradigm,
>>new paradigm, Newton, and so forth does nothing to make any point you want to
>>make.
>>
>>
>>
>>
>>>
>>>GT evaluates the chances to pull you into chaos of a heavy king attack.
>>>this is the fog where the quiesence search has problems to evaluate and to
>>>handle the position.
>>
>>
>>It doesn't do this. It follows the path its search, its q-search, and its
>>evaluation says is optimal according to the terms in the evaluation. It
>>doesn't know squat about "into the fog". That is nonsense. It is _still_
>>a normal "bean-counter" type program. The shape/size of the beans is different,
>>but that is _all_ that is different. You can try to cast it in the image of
>>CSTal if you want. But it isn't.
>>
>
>So what is your take? I never heard it spelt out in one place without any
>flag-waving. Do you say *all* programs are bean-counters except CSTal? Or? It's
>a bean-counter too? Tell your theory of what is what and why and where. Is there
>one school, two, three? One paradigm, none, two, three? Kuhn's theory nonsense
>or sense? What? Tell us. Tell us, o mighty one, what is your answer?
I believe that _most_ programs, built around the alpha/beta framework, would
fit into what you want to call "bean counters". The only problem with your
"painting with a broad brush" is that we all count things that may or may not
look like 'beans'.
In your case, I am not sure what you do since I don't have a copy of your
program, I don't know anybody that does, and I wouldn't have time to study
it in any case. In the case of other programs, I play them all the time on
ICC, and the operators provide ongoing analysis from the programs, just as
I have crafty do when they ask. As a result, I develop a pretty good feel
for what the program is doing and what it isn't. Hiarcs is very good in its
evaluation _and_ in its search extensions. But it doesn't do fog. I don't
believe _anybody_ does "fog" regardless of the hyperbole and hand-waving.
In the case of Gambit Tiger, I have seens its PVs, and its scores. And in
looking at the PV, and the score, it is pretty easy to see where the score is
coming from. It _really_ likes open files with rooks/queens on them. Even
if the file is blocked forever by a bishop or knight. When there is no piece
closing the file, it is often right. When there is such a piece, it is often
wrong. But it isn't doing any deep, selective search, trying to find a bit of
"fog" to hide in. It is just evaluating things that are generally known to be
good in attacking, but missing a few things that are also known to be good to
counteract such an attack.
I'm sure it will get better. I am sure the rest of us will as well. It is
just another program, with a slightly different eval than most. If you want a
new paradigm, look at p.conners. That _is_ different. _totally_ different.
But alpha/beta with endpoint evals has been done forever. Speculative evals
have been done forever. The game discussed here wasn't resolved by any wild
kingside attack. It ended up in an endgame.
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.