Author: Robert Hyatt
Date: 17:24:58 10/13/98
Go up one level in this thread
On October 13, 1998 at 17:44:15, blass uri wrote:
>
>On October 13, 1998 at 17:40:22, Robert Hyatt wrote:
>
>>On October 13, 1998 at 17:01:51, Moritz Berger wrote:
>>
>>>On October 13, 1998 at 15:56:09, Robert Hyatt wrote:
>>>>On October 13, 1998 at 13:04:02, Moritz Berger wrote:
>>>>
>>>>just so you know, some of these are openings Crafty is *not* going to
>>>>play very well, most likely, unless the "cramped pawns and blocked
>>>>position" code is working better than I thought. :)
>>>
>>>So far, Crafty has been doing quite well (50%), despite some games with wrong
>>>settings for Crafty (which will be repeated after the 10th Nunn position). That
>>>some openings are not well suited for computers is intended in the Nunn test as
>>>far as I understand it - measuring progress where it is needed most seems to be
>>>a good thing for such a test.
>>>
>>>>this is one example, although I haven't looked at the precise variation to
>>>>see how the position starts off... and how did crafty draw the first game
>>>>with an eval of -6.something? that ought to be lost? Could the result
>>>>tag be wrong somehow?
>>>
>>>In the Fritzian universe, eval -xx.yy means: advantage for black (as opposed to
>>>eval xx.yy = good for white). Crafty scores are parsed by Fritz, so
>>>
>>>{0.05/12 428}
>>>
>>>translates to:
>>>- 0.05 advantage for white
>>>- eval at 12 ply
>>>- after 428 seconds, not taking into account time spent on that position in
>>>permanent brain
>>>
>>>Position at move 73. (white to move) in the drawn game
>>>(FEN: 8/4k1K1/6r1/4n3/8/8/7R/8 w - - 0 73) is a draw per KT-EGTB.
>>>Indeed, after move 116 we still have 5R2/6r1/7K/4n3/4k3/8/8/8 b - - 0 116 and no
>>>progress has been made. So I adjudicated it a draw.
>>>
>>>Why the -6.something score from Crafty? Crafty 15.19 engine for Fritz uses
>>>different opening book and EGTB mechanisms than wcrafty or linux-crafty:
>>>
>>>- The Fritz UI controls the opening book. When it leaves the book, it hands over
>>>to the active engine (Hiarcs, Junior, Fritz, Crafty, Exchess, Nimzo,
>>>CD-Endgame).
>>>
>>>- per move timing is *not* controlled by the Fritz UI but by the engines
>>>themselves - they are initialized with proper time controls at the beginning of
>>>each game, but then left alone.
>>>
>>>- Endgames are either done by selected "playing" engines or "CD-Endgame" engine
>>>(that is selected automatically if you select that option (default) and only
>>>probes some 5 man KT-EGTB at the root).
>>>
>>>Obviously, there's no crafty.rc or any other way to pass parameters to the
>>>Crafty engine other than using the engine-parameters dialog with options offered
>>>by the UI, namely (for Crafty 15.19): "no tricks" and "draw score normal".
>>>
>>>I remember reading a statement from Bob that Crafty would suffer if it didn't
>>>have it's own book. I have played several dozens of Blitz (60/5) games
>>>Crafty15.19/PII-450 vs. Hiarcs6/P233MMX and Crafty scored about 50% with the
>>>FritzPower book.
>>>
>>>Why the "good" result? The reason is simply that the Fritz PowerBook doesn't
>>>contain any Fritz-specific weightings but only move frequency, performance
>>>percentage and ELO-performance plus learning-adjusted weightings (learning is
>>>also not done by the engine but by the UI, using engine evals for guidance). So
>>>the Fritz PowerBook concept works equally well for all engines because the
>>>learning-values are all 0 before the 1st game.
>>>
>>>Do you have any more questions I didn't answer?
>>>
>>>I feel that if Crafty gets 50% in the Nunn test or in Blitz games against Hiarcs
>>>and if Crafty has a PII-450 vs. P233MMX for its opponent, this still shows that
>>>Crafty is very likely not weaker than 100 SSDF-ELO than the top commercial
>>>programs - after all, as Fritz engine it doesn't get it's own book (doesn't
>>>matter in the Nunn test, of course), it doesn't get full EGTB support and
>>>doesn't have its own positional learning. I think the result shows that Crafty
>>>is a serious, strong chess program and this is exactly the reason why I also
>>>enjoy it for consultation in ChessBase 7, not only as an opponent to play
>>>against. Remember how "well" another program recently did against Hiarcs 6 in
>>>the Nunn test with the same hardware handicaps against Hiarcs ...
>>>
>>>Moritz
>>
>>ok... the - eval now makes sense... This is the way crafty normally annotates
>>a game on its own, too, but I didn't stop to look at a position... although I
>>now need to look at the game to see why it couldn't win a +6 ending...
>
>It could not win a +6 ending because there is no basis to +6
>The problem is with the evaluation function.
>
>Uri
I would probably choose to disagree here... the +6 is *not* a positional
score, as you wrongly suggested. It is a *tactical* score. Ie in the
PV I have for this position, there is a definite pawn promotion at the
end... I'm not sure what I'm overlooking, but in the position at the
end of the pgn file, Ra7+ Kg6 Rg6+ seems to be a real problem unless I
miss something?
Bob
>>
>>And I wasn't questioning what you are doing. But some openings don't work. For
>>some programs...
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.