Computer Chess Club Archives




Subject: Re: Let's analyze move 36

Author: Chris Whittington

Date: 12:21:16 10/08/97

Go up one level in this thread

On October 08, 1997 at 13:52:59, Amir Ban wrote:

>On October 08, 1997 at 10:29:24, Chris Whittington wrote:
>>On October 08, 1997 at 07:08:01, Amir Ban wrote:
>>>>>>>>>Let's try to analyze the critical variation in DB-GK game 2. The move is
>>>>>>>>>36 (axb5 was played, don't confuse it with move 37).
>>>>>>>>>The variation to look at is after 36. Qb6 Qe7 37. axb5 Rab8 38. Qxa6 e4
>>>>>>>>>39. Bxe4 Qe5. What does Black have here ?
>>>>>>>>>At ply (or whatever) 10, DB gives this PV:
>>>>>>>>>Qb6 Qe7 axb5 Rab8 Qxa6 e4 Bxe4 Qe5 Bf3 (at lower depths it tries g4) Bf3
>>>>>>>>>Rd8 Qa7 Qxc3 Bh5 +0.74.
>>>>>>>>>At ply 11, there is no PV and the eval is +0.48.
>>>>>>>>>This variation was mentioned by Seirawan in his analysis, and the GK
>>>>>>>>>team used Hiarc & Fritz to analyze it to death.
>>Wrong choices :(
>>>>>>>>>Junior, by the way, likes an intermediate 37...a5, which creates the
>>>>>>>>>same position but with Ra2 misplaced.
>>>>>>>>>What do you make of this ?
>>CSTal always goes for Qe5, so I played this and let white think on a
>>PV: Bf3 Rd8 Rc2 Qf5 Qa2 Rxb5 Qe2 Qf4 Qa7 +0.44
>Phew. Good job ! I'll be watching you in Paris.
>>1. Note how white has to pull his pieces back to defend
>>2. King safety (Tal) function fluctuates between -1.00 and -2.50 pawns
>>from start to end of PV
>>3. CSTal and DB appear to agree on the evaulation +0.48 or +0.44
>>4. Therefore DB, in this game, is using a manic positional king safety
>>evaluation function, similar to CSTal
>>5. The operator, if he had much understanding of chess, would have seen
>>the deeper PV's coming from DB, seen the king attack possibilities for
>>black; and, as a human, realised that DB was in deep shit after Qb6,
>>white pieces scattered, kingside opened up, opposite bishop+queen
>>attack, potential for rooks to get involved, while white just has some
>>distant freepawns as compensation. I imagine he would have breathed a
>>sign of relief after axb5 appeared on his console; axb5 trades the rooks
>>and black has less potential to then hit the white king.
>Quite right, except that Kasparov has some more sinister theories about
>what actually happened. I can understand why. It seems an exclusively
>human choice to prefer throwing some pawns and seize the initiative
>(even though this may lose) rather than defend passively (even though
>this may draw). Computers are  supposed to blindly add evaluation terms
>and not know anything about this.
>Black still does not win here by any long shot, and Kasparov didn't say
>so. The attack is speculative and fits CSTal well. I guess you were
>joking when you said DB had your manic eval function, but seriously, it
>does not make sense for them to do anything speculative in view of their
>other strengths, and if they do, their huge NPS may actually work
>against them.

No, I jest not. its the obvious thing to do. What do you think the hired
GM's said to the DB programmers ?

"If you play material + depth, he'll tie you up in knots like game 6,
first match"

"Steer towards king attack complexities; that plus 15-30 ply or so is
too much for any human to cope with". That would have been my advice

>>>Yes, thank you. But it would be even more interesting to follow the
>>>variation Qb6 Qe7 axb5 Rab8 Qxa6 e4 Bxe4 Qe5 and look what you find
>>>Remember this is what DB was thinking about, actually beginning at the
>>>first iteration it does (ply 8) after 1 second ! With Black giving up 3
>>>pawns (actually 2 since one is regained easily), it evaluates it as
>>>+0.74 on ply 10, finally +0.48 on ply 11 (see my original post). It's a
>>>remarkable choice, but what is it based on ?
>>>If it's tactical, a PC program starting with Qe5 (8 plies less to
>>>calculate) should find it. Do you ? If it's positional, your program is
>>>known to have good judgement in this kind of position.
>>>Kasparov tried it with Hiarcs and didn't see the explanation. Hiarcs
>>>doesn't come even near to the DB evaluation. I told him DB's eval must
>>>have been positional, and that they probably had very high penalties for
>>>open king positions. His answer: "Then how could it play moves like g5
>>>in the 1st game and b4 in the 4th ?".
>>Interesting point.
>>1. The DB king attack function is inaccurate, broken and sees some
>>things only
>>2. Its very accurate and doesn't mind broken king positions that are
>>3. They used versions of DB with different (seriously different)
>>evaluation weightings to confuse Kasparov.
>>I prefer option 3. Its well known (well to me anyway) that one of the
>>steps from IM to GM and SuperGM is an ability to get into the mind of
>>the opponent, to know what it is that he knows, to predict the sort of
>>things he will do. You can see Kasparov continually worrying about just
>>this during the match and after. Keep changing the parameters, and you
>>badly confuse a SuperGM.
>Too clever by half, I think. Not to mention that such violent changes in
>eval may easily lose games.

yes, but. You are a programmer speaking. programmers onyl have one
version, the last one, which, of course is the best one. SO you believe
in one version.

Operators have many versions, serval books, learnt data. You shoudl hear
operators thinking. "I have version 93 and version 240 and and and, and
this book and that book, and I use this version against this opponent
and and and "

Operators and advisors can only influence the result with their
versions. So they use their versions and switch between them. Nothing
unusual. Happens all the time.

>Incidentally, up to ply 10 DB finds g4 of all moves as the right answer
>to Qe5. What does your manic king safety think about that ?

I'm leaving Bf3 on overnight.

Will try this one tomorrow



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.