Author: Robert Hyatt
Date: 18:52:32 01/04/03
Go up one level in this thread
On January 04, 2003 at 21:41:31, scott farrell wrote:
>On January 04, 2003 at 15:01:34, Robert Hyatt wrote:
>
>Yes it is about the knight on a8. A human doesnt need a horizon to see the
>knight on a8 should be scored way down, and will probably be lost, but most
>programs if it cant see it being lost inthe qsearch, consider the piece to be
>safe.
Maybe. Crafty has code to hate a knight on a8, and the PV it had showed the
knight getting back out quickly. Here is the PV:
15 1:44 0.96 2. Nxb6 Nxd3 3. Nxa8 Nf4 4. Rfe1 Nf6
5. Re5 b6 6. Nc7+ Kd7 7. Na6 Bd6 8.
Re3 N4xh5 9. Rae1 Nxg3 10. fxg3
As you see, in that variation, the knight gets out pretty quickly. The
only issue is does white have time to do that, or does white have to handle
some other threat by black that prevents getting the knight out? That is not
easy to compute for me (as a human) or crafty (as a computer).
My gut reaction is that this is winning for _white_... not black... if white
plays well. If white doesn't realize that the knight is dangerously close to
being lost on a8 and take immediate action to extricate it, I could easily see
a program losing that knight by a misstep here or there...
>
>Thanks to bit boards, I check the mobility of knights/bishops/rooks - and I
>think it paid off big time here.
>
>The knight gets royally stuck on a8, and my chompster scores a knight with no
>move, other than into pawns and pawns attacks, as less than a pawn, and a knight
>with only 1 move as about half a knight.
>
>The knight on a8 has one move, but it too is trapped, and only leads back to a8.
Maybe or maybe not. See my PV above which gets out thru c7 (with check) and
a6.
>
>Dorky's score moved down very rapidly with each move in the game, but didnt
>really take the toll of the knight into account until actually capturing it was
>within its horizon of 13-14 ply. Then it played all sorts of strange moves to
>try to prevent it, as it didnt understand that it was undefendable because it
>couldnt move.
>
>
>>On January 04, 2003 at 14:39:40, scott farrell wrote:
>>
>>>This position occured in a game chompster vs dorky on ICC (45 10 time control
>>>practices for CCT5).
>>>
>>>[D]r3kb1r/pp1n1pp1/1qp1p2p/3n3P/2NP1B2/3Q2N1/PPP2PP1/R4RK1 b kq - 0 14
>>>
>>>My engine says NxB is slightly winning for black, it failed high big time at
>>>only depth 8 !!!
>>>sfarrellc(C) whispers: , D10/11(11): 0.29 : D5F4 , 25.176/48.0 secs
>>>
>>>Crafty says it is +3 winning for white.
>>>Dorky(C) whispers: score=+2.61 depth=13 Knps=111 cpu=99% time=87.05s
>>>
>>>Here is crafty's annotation:
>>>({9:+2.92} 14. ... Nxf4 15. Nxb6 Nxd3 16. Nxa8 Nxb2 17. Rab1 Ba3 18. Nc7+ Kd8
>>>19. Na8 +-)
>>>({9:+1.04} 14. ... Qd8 15. Bd2 N7b6 16. Ne5 Bd6 17. Rfe1 Qh4 18. Qe4 Qxe4 19.
>>>Rxe4 Bxe5 20. Rxe5 +/-)
>>>({9:+1.52} 14. ... Qb5 15. Bd6 Nb4 16. Qe2 Nb6 17. Bxf8 Kxf8 18. b3 Nxc4 19.
>>>bxc4 Qg5 +-)
>>>({9:+1.66} 14. ... Qa6 15. Nd6+ Bxd6 16. Qxa6 bxa6 17. Bxd6 N7f6 18. Rfe1 Rd8
>>>19. Bc5 Nf4 +-)
>>>({9:+2.03} 14. ... Qb4 15. Bd2 Qa4 16. Ne4 Qa6 17. Ncd6+ Bxd6 18. Nxd6+ Ke7 19.
>>>Nc4 Qb5 +-)
>>>
>>>
>>>NxB is a clear winner !!!! It was an easy win from there.
>>>
>>>How many other engines see this? It took 19 more moves to actually recover the
>>>material that my engine could see coming !!!!
>>
>>
>>
>>
>> Crafty thinks Qd8 and Nxf4 are different in score, but not +3. If I
>>let crafty think, it thinks black is better with a score of -.9 after Qd8.
>>If I play Nxf4, it thinks white is better with a score of +1.16 and would play
>>Nxb6 thru depth 14, which only takes 10 seconds or so to reach.
>>
>>I haven't looked at it, but I am not sure that Nxf4 "is a clear winner"
>>If this is about the white knight getting stuck at a8, it _might_ be a
>>problem, but I doubt that any program will see it from the starting position.
>>
>>I can't get the score you posted above so I don't know what is going on there.
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.