Author: Uri Blass
Date: 08:46:09 10/08/01
Go up one level in this thread
On October 08, 2001 at 11:24:10, Robert Hyatt wrote: >On October 08, 2001 at 09:31:57, Uri Blass wrote: > >>On October 08, 2001 at 07:29:04, Bernhard Bauer wrote: >> >>>On October 07, 2001 at 21:39:52, Robert Hyatt wrote: >>> >>>> >>>>2. You can tone down (or even disable) the null-move search. You can >>>>try sel=2/3 (the default) then sel=2/2, sel=1/1, and finally sel=0/0 which >>>>will turn null-move off. This will further reduce the search depth and overall >>>>strength. >>>> >>> >>>From my experience sel=0/0 doesn't turn null-move completely off. >>>Here an example with sel=0/0 >>> >>>[D]4B/8/6N/5p/1r4p/6pk/7b/4K2Q w - - >>> >>> >>> (4) 6-> 1.90 -0.44 1. Bc6 Rb1+ 2. Ke2 Rxh1 3. Nf4+ Kh4 >>> 4. Bxh1 Kg5 5. Ke3 >>> (3) 7 1.96 -0.22 1. Bc6 Rb1+ 2. Ke2 Rxh1 3. Nf4+ Kh4 >>> 4. Bxh1 Bg1 5. Bg2 Bd4 >>> 7 2.75 ++ 1. Kd2!! >>> (2) 7 3.79 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 >>> (2) 7-> 6.29 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 >>> 8 8.14 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 >>> (2) 8-> 14.57 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 >>> 9 20.50 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 10. Nf4 >>> (2) 9-> 42.50 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 10. Nf4 >>> 10 1:05 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 10. Nf4 Bxf4 >>> (2) 10-> 2:22 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 10. Nf4 Bxf4 >>> 11 3:35 0.00 1. Kd2 g2 2. Qe1 g1=Q 3. Qh4+ Kg2 4. >>> Bc6+ Re4 5. Bxe4+ fxe4 6. Qxg4+ Kh1 >>> 7. Qxe4+ Qg2+ 8. Ke3 Qxe4+ 9. Kxe4 >>> Bg3 10. Nf4 Bxf4 11. Kxf4 Kh2 >>> time=5:00 cpu=201% mat=4 n=142525014 fh=89% nps=474k >>> ext-> chk=10823921 cap=287226 pp=207595 1rep=668532 mate=112510 >>> predicted=0 nodes=142525014 evals=28834988 >>> endgame tablebase-> probes done=0 successful=0 >>> hashing-> trans/ref=47% pawn=99% used=99% >>> SMP-> split=1018 stop=90 data=7/32 cpu=10:05 elap=5:00 >>> >>> >>>Kind regards >>>Bernhard >> >>I am not sure if the problem here is null move pruning. >> >>The problem may be that programs need to know that there is a chance to win in >>KN vs something(otherwise they may evaluate it as not more than a draw and stop >>to search) >> >>There are programs without this knowledge so they cannot see that >>1.Bc6 Rb1+ 2.Ke2 Rxh1 3.Bg2+ is leading to a forced mate. >> >>Deep Fritz with the default parameters(null move pruning) has no problem to find >>the win because it has this knowledge that king and knight can mate. >> >>Uri > > > >I think your statement is backward. It should read > >"deep fritz has no problem to find the win because it _lacks_ the knowledge >that in 99.99999% of the cases, a king + knight can not mate." This is a >decision made based on speed. And it is a decision that will let the thing >trade away it's last pawn, thinking it will still be +3. You are right that it is Deep Fritz lack the knowledge but I did not talk about knowledge in the evaluation. I thought about the knowledge not to stop to search and not knowledge in the evaluation. I agree that programs do not need knowledge not to stop to search and need knowledge to stop to search so you are right that I could say thart deep Fritz does not have the knowledge. Junior7 simply stops to search in positions that are draws in 99.9999% of the cases. It simply knows that king and knight cannot mate and it even does not check for checkmates except the first ply. Here is some evaluation of Junior after learning from previous search when it can see the score of mate in 1 only after stopping to learn from the hash tables New position 8/8/8/5p2/5Np1/8/6pb/4K1kr w - - 0 1 Analysis by Junior 7: 6.Nxg2 Kxg2+ 7.Ke2 g3 -+ (-8.67) Depth: 3 00:00:00 6.Ne6 Bg3+ 7.Ke2 f4 -+ (-7.94) Depth: 3 00:00:00 6.Ne2# = (0.00) Depth: 3 00:00:00 6.Ne2# = (0.00) Depth: 6 00:00:00 6.Ne2# = (0.00) Depth: 9 00:00:00 6.Ne2# = (0.00) Depth: 12 00:00:00 1kN 6.Ne2#! +- (#1) Depth: 14 00:00:01 7kN 6.Ne2#! +- (#1) Depth: 14 00:00:01 7kN Note that I checked that in this case Junior can find all the right moves in the position because the alternative is not better than a draw but simply does not see a winning score. 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.