Author: Bruce Moreland
Date: 19:29:23 06/05/00
Go up one level in this thread
On June 05, 2000 at 20:25:20, John Merlino wrote: >On June 05, 2000 at 17:51:35, blass uri wrote: > >>On June 05, 2000 at 11:46:41, Robert Hyatt wrote: >> >>>On June 05, 2000 at 10:31:32, blass uri wrote: >>> >>>>On June 05, 2000 at 09:09:51, Robert Hyatt wrote: >>>> >>>>>On June 05, 2000 at 08:22:44, Steffen Jakob wrote: >>>>> >>>>>>Hi! >>>>>> >>>>>>This is a position from an interesting standard game MissSilicon - Hossa, played >>>>>>today at ICC: >>>>>> >>>>>>[D]5k2/7K/6P1/1p3p2/1P5P/1Pb5/8/8 w >>>>> >>>>> >>>>> >>>>>I think there are times where you have to depend on your search. Normally >>>>>this is won by black, because black has a bishop to stop white's pawns. I >>>>>suspect that if you try to write special-purpose code to catch this, it will >>>>>end up being wrong more than it is right. >>>> >>>>I agree that writing a special code to catch this case and many other cases is >>>>not trivial but I believe that it is not impossible. >>>> >>>> This is a precise tempo-counting >>>>>issue that just barely lets the 'loser' win. >>>>> >>>>>I'd likely just take the loss and run. It takes Crafty 9 plies (0 seconds >>>>>of course) to see that the bishop is not winning. I don't see an obvious >>>>>evaluation trick to make this show up faster... >>>> >>>>Hiarcs7.32 needs only 5 plies to see that white is winning because of >>>>extensions. >>>> >>>>Uri >>> >>> >>>That doesn't matter. How _long_ does Hiarcs need? It doesn't matter whether >>>you need 5 plies or 9 plies. What matters is "how many seconds"??? Because it >>>is easy to extend a lot and pick this up quicker (shallower plies) but take >>>longer overall to find the problem. Here is what I get, for reference: >>> >>> nss depth time score variation (1) >>>starting thread 1 >>>starting thread 2 >>>starting thread 3 >>> 1 0.00 -5.90 1. h5 >>> 1-> 0.00 -5.90 1. h56 >>> 2 0.00 -- 1. h5 >>> 2 0.00 -6.77 1. h5 Bxb4 >>> 2 0.00 -6.52 1. Kh6 f4 >>> 2-> 0.01 -6.52 1. Kh6 f4 >>> 3 0.01 -6.22 1. Kh6 f4 2. h5 >>> 3 0.01 ++ 1. h5!! >>> 3 0.01 -5.40 1. h5 Bg7 2. h6 >>> 3-> 0.01 -5.40 1. h5 Bg7 2. h6 >>> 4 0.02 -5.14 1. h5 f4 2. h6 Bxb4 >>> 4-> 0.02 -5.14 1. h5 f4 2. h6 Bxb4 >>> 5 0.23 ++ 1. h5!! >>> 5 0.24 -3.95 1. h5 f4 2. h6 f3 3. g7+ Bxg7 4. hxg7+ >>> 5-> 0.24 -3.95 1. h5 f4 2. h6 f3 3. g7+ Bxg7 4. hxg7+ >>> 6 0.25 -4.26 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 f4 >>> 6-> 0.26 -4.26 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 f4 >>> 7 0.27 -4.26 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 f4 4. >>> Kh7 >>> 7-> 0.27 -4.26 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 f4 4. >>> Kh7 >>> 8 0.28 -4.43 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 f4 4. >>> g7+ Kg8 5. Kg5 >>> 8-> 0.29 -4.43 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 f4 4. >>> g7+ Kg8 5. Kg5 >>> 9 0.38 ++ 1. h5!! >>> 9 0.49 3.78 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 Kg7 5. Kxf5 >>> 9-> 0.49 3.78 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 Kg7 5. Kxf5 >>> 10 0.50 3.68 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kg7 >>> 10-> 0.51 3.68 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kg7 >>> 11 0.52 3.78 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kg7 6. Kf5 >>> 11-> 0.53 3.78 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kg7 6. Kf5 >>> 12 0.55 3.88 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kh8 6. Ke5 Kg7 >>> (2) 12-> 0.58 3.88 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kh8 6. Ke5 Kg7 >>> 13 0.60 3.78 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kh8 6. Ke5 Kg7 7. Kf5 >>> (2) 13-> 0.63 3.78 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kh8 6. Ke5 Kg7 7. Kf5 >>> 14 0.66 ++ 1. h5!! >>> 14 6.49 4.19 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kf8 6. Kg5 Ke7 7. g7 >>> Kf7 8. Kh6 >>> 14-> 6.56 4.19 1. h5 Bg7 2. h6 Bxh6 3. Kxh6 Kg8 4. >>> Kg5 f4 5. Kxf4 Kf8 6. Kg5 Ke7 7. g7 >>> Kf7 8. Kh6 >>> >>> >>>Crafty finds the right move from depth=1. After .38 seconds it realizes that >>>white is winning. I can crank up the passed pawn extension and see this >>>quicker in terms of depth, but the time will probably be slower overall. IE >>>don't be mislead by 'shallow depth finding'. I think depth doesn't matter at >>>all. It is _time_. >> >>I do not know the exact time but it needs clearly less than 1 second on p200MMX >>to see evaluation of +4.88 >> >>Uri > >Chessmaster 7000 is the same, needing less than 1 second to find the move with a >score of +4.44 at a depth of 9/10. > >1.h5 Bg7 2.h6 Bxh6 3.Kxh6 Kg8 4.Kg5 Kf8 5.Kxf5 -- the same line that Crafty >finds. So the results are pretty much the same. > >The score jumps dramatically to +8.43 at depth 12/13 after 13 seconds. > >5... Kg8 6.Ke6 Kg7 7.Kd7 Kxg6 8.Kc6 Kf5 9.Kxb5 > >jm I would like to gently point out that none of these means anything. You guys are arguing about rounding error, and in the final position everything I've seen is just a heuristic score that is only coincidentally accurate. You can all get to a position where white is up a pawn. That's not hard. Try the position that I posted elsewhere in this thread and see how long it takes you to return a draw score without KPP vs KP. bruce
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.