WAC 109

Programming Topics (Computer Chess) and technical aspects as test techniques, book building, program tuning etc

Moderator: Andres Valverde

WAC 109

Postby Robert Pope » 17 Jun 2008, 14:40

[diag]rn2k1nr/pbp2ppp/3q4/1p2N3/2p5/QP6/PB1PPPPP/R3KB1R b KQkq -[/diag]
Best move: c3

My engine can never solve WAC 109, but Crafty solves it in 20K nodes! Can someone explain the key concepts that make this the right move? I assume it must be evaluation-related to be found so quickly when Beaches can't find it at all.

Crafty's PV is: . ... c3 2. Qxd6 cxd6 3. Bxc3 dxe5 4. Bxe5
for a 1 pawn gain for black. But to my untrained eye, it looks like black is exchanging Q/N/P and only getting Q/N in return, so white is gaining ground. Why isn't black better off going for a straight QxQ exchange, or cxb3 to keep the material exchange even?
Robert Pope
 
Posts: 39
Joined: 08 Apr 2006, 17:27

Re: WAC 109

Postby Teemu Pudas » 17 Jun 2008, 14:50

But to my untrained eye, it looks like black is exchanging Q/N/P and only getting Q/N in return, so white is gaining ground.

Count again. White loses a knight for two pawns.
Teemu Pudas
 
Posts: 124
Joined: 16 Apr 2007, 14:03

Re: WAC 109

Postby Robert Pope » 17 Jun 2008, 15:20

I am so not visual. :| So it is a straight material question. I'll have to trace and see why that branch keeps getting pruned then. Thanks!
Robert Pope
 
Posts: 39
Joined: 08 Apr 2006, 17:27

Re: WAC 109

Postby Dann Corbit » 17 Jun 2008, 22:16

Robert Pope wrote:[diag]rn2k1nr/pbp2ppp/3q4/1p2N3/2p5/QP6/PB1PPPPP/R3KB1R b KQkq -[/diag]
Best move: c3

My engine can never solve WAC 109, but Crafty solves it in 20K nodes! Can someone explain the key concepts that make this the right move? I assume it must be evaluation-related to be found so quickly when Beaches can't find it at all.

Crafty's PV is: . ... c3 2. Qxd6 cxd6 3. Bxc3 dxe5 4. Bxe5
for a 1 pawn gain for black. But to my untrained eye, it looks like black is exchanging Q/N/P and only getting Q/N in return, so white is gaining ground. Why isn't black better off going for a straight QxQ exchange, or cxb3 to keep the material exchange even?


This is not a position with any known alternate moves.
Here is Rybka's analysis:
Code: Select all
Analysis from Q:\epd\sub\wac109.epd   
6/17/2008 12:39:35 PM Level: 3600 Seconds
Analyzing engine: Rybkav2.3.2a.mp.x64

1) c3; id WAC.109;     
    Searching move: c4-c3
    Best move (Rybkav2.3.2a.mp.x64): c4-c3
    identical moves! Found in: 00:00
      5   00:00          3.618   217.931   -0.04   c4c3 d2xc3
      6   00:00          5.671   341.594   -0.05   c4c3 d2xc3 Qd6xe5 c3c4
      7   00:00         18.233   388.970   +0.31   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7
      8   00:00         36.043   467.190   +0.41   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3a5
      9   00:00         48.261   520.202   +0.38   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3a5 Nb8c6
     10   00:00         88.201   522.068   +0.47   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3a5 Nb8c6 Qa5xb5
     11   00:00        132.635   541.108   +0.44   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 Nb8d7 Ra1c1 000
     12   00:00        201.920   549.909   +0.51   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 Nb8d7 Ra1c1 000 Bb2xg7 Rh8g8
     13   00:00        288.845   555.972   +0.51   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 Nb8d7 Ra1c1 000 Bb2xg7 Rh8g8 Bg7d4
     14   00:01        638.978   590.003   +0.56   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 Nb8d7 Bb2xg7 Rh8g8 Bg7d4 Ne7f5 Bd4c3
     15   00:01      1.005.485   599.660   +0.56   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 000 Rf8c8 e2e4 c7c6 b5xc6
     16   00:05      3.085.154   595.513   +0.59   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 Nb8d7 Ra1c1 Ra8c8 g2g3
     17   00:07      4.305.376   605.092   +0.57   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 Nb8d7 Ra1c1 Ra8c8 g2g3
     18   00:11      6.540.957   608.074   +0.56   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 a7a6 e2e4 a6xb5 Bf1xb5
     19   00:17      9.909.024   596.697   +0.53   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 a7a6 e2e4 a6xb5 Bf1xb5
     20   00:30     17.960.637   603.026   +0.55   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 a7a6 e2e4 a6xb5 Bf1xb5
     21   00:57     33.835.561   599.771   +0.54   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 a7a6 e2e4 a6xb5 Bf1xb5
     22   02:31     89.350.438   604.014   +0.60   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 Rf8d8 e2e4 c7c6 Ra1c1
     23   05:06    182.640.384   610.291   +0.60   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 Rf8d8 e2e4 c7c6 Ra1c1
     24   09:00    319.262.530   605.278   +0.56   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 Rf8d8 e2e4 c7c6 Ra1c1
     25   18:31    647.804.172   597.087   +0.60   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 Rf8d8 e2e4 c7c6 Ra1c1
     26   33:25   1.171.048.287   597.987   +0.60   c4c3 d2xc3 Qd6xe5 c3c4 Qe5e7 Qa3xe7+ Ng8xe7 c4xb5 00 f2f3 a7a6 e2e4 a6xb5 Bf1xb5
   6/17/2008 1:39:38 PM, Time for this analysis: 01:00:00, Rated time: 00:00

1 of 1 matching moves
6/17/2008 1:39:39 PM, Total time: 1:00:03 AM
Rated time: 00:00 = 0 Seconds


[D]rn2k1nr/pbp2ppp/3q4/1p2N3/2p5/QP6/PB1PPPPP/R3KB1R b KQkq - bm c3; id WAC.109;
[D]rn2k1nr/pbp2ppp/3q4/1p2N3/8/QPp5/PB1PPPPP/R3KB1R w KQkq -
[D]rn2k1nr/pbp2ppp/3q4/1p2N3/8/QPP5/PB2PPPP/R3KB1R b KQkq -
[D]rn2k1nr/pbp2ppp/8/1p2q3/8/QPP5/PB2PPPP/R3KB1R w KQkq -
[D]rn2k1nr/pbp2ppp/8/1p2q3/2P5/QP6/PB2PPPP/R3KB1R b KQkq -
[D]rn2k1nr/pbp1qppp/8/1p6/2P5/QP6/PB2PPPP/R3KB1R w KQkq -
[D]rn2k1nr/pbp1Qppp/8/1p6/2P5/1P6/PB2PPPP/R3KB1R b KQkq -
[D]rn2k2r/pbp1nppp/8/1P6/8/1P6/PB2PPPP/R3KB1R b KQkq -
[D]rn3rk1/pbp1nppp/8/1P6/8/1P6/PB2PPPP/R3KB1R w KQ -
[D]rn3rk1/pbp1nppp/8/1P6/8/1P3P2/PB2P1PP/R3KB1R b KQ -
[D]rn3rk1/1bp1nppp/p7/1P6/8/1P3P2/PB2P1PP/R3KB1R w KQ -
[D]rn3rk1/1bp1nppp/p7/1P6/4P3/1P3P2/PB4PP/R3KB1R b KQ -
[D]rn3rk1/1bp1nppp/8/1p6/4P3/1P3P2/PB4PP/R3KB1R w KQ -
[D]rn3rk1/1bp1nppp/8/1B6/4P3/1P3P2/PB4PP/R3K2R b KQ -
Dann Corbit
 

Re: WAC 109

Postby Vladimir Medvedev » 18 Jun 2008, 12:21

GreKo 5.7.1:

Code: Select all
rn2k1nr/pbp2ppp/3q4/1p2N3/2p5/QP6/PB1PPPPP/R3KB1R b KQkq - 0 1

  1       -57        0           299    1. ... Bd5
  2       -57        1          1723    1. ... c3 2. Qxd6 cxd6
  3       -52        1          3774    1. ... f6 2. Qxd6 cxd6
  4        -2        3          9586    1. ... c3!
  4       103        4         12379    1. ... c3 2. Qxd6 cxd6 3. Bxc3 dxe5 4. Bxe5
User avatar
Vladimir Medvedev
 
Posts: 129
Joined: 29 Sep 2004, 10:03
Location: Moscow, Russia

Re: WAC 109

Postby Robert Pope » 18 Jun 2008, 15:36

Okay, I broke down the evaluation before and after the winning sequence of moves. It looks like my pawns are valued too strongly for starters:

1. The lost white knight was worth 312 (300+12 positional bonus).
2. The black pawn on c4 was worth 160 (100+20 rank bonus + 40 pawn structure bonus).
3. The black pawn of c7 was worth 130 (100 + 5 rank bonus + 25 pawn structure bonus).

So net material gain was only 22 centipawns after the exchange.

4. The white bishop got a 10 point bonus for better placement after the exchange
5. White's relative mobility increased by 4, giving another 16 point bonus for white.

So after all that, Beaches only saw a 6 centipawn improvement from the exchange and thought there were better options.

The only way I can get that kind of pawn structure bonus is from Beaches evaluating both pawns as passed, which they aren't, so that must be a bug in my eval. If I fix that, then the trade will come out as a 3/4 pawn gain for black and that should be enough to accept the trade.

ETA: My 1/4 pawn "passed pawn" bonus is really just a "no pawn in my direct path" bonus, so that definitely needs cleaning up.
Robert Pope
 
Posts: 39
Joined: 08 Apr 2006, 17:27

Re: WAC 109

Postby Dann Corbit » 18 Jun 2008, 19:10

Robert Pope wrote:{snip}
ETA: My 1/4 pawn "passed pawn" bonus is really just a "no pawn in my direct path" bonus, so that definitely needs cleaning up.


The Beowulf code uses:
dangerous passed pawn bonus
very dangerous passed pawn bonus
unstoppable passed pawn bonus
{and other pawn factors}

Here are the pawn related factors:
Code: Select all
/* Pawn Location */

#define DEFENDED_PAWN

#define DOUBLED_PAWNS
#define DOUBLED_PAWNS_ISO
#define HOSTILE_BLOCKADE
#define PIECE_BLOCKADE
#define ISOLATED_PAWN

#define BACKWARD_PAWN_1
#define BACKWARD_PAWN_2
#define BACKWARD_PAWN_3

#define UNSTOPPABLE_PP
#define KING_SAFETY
#define KING_TROPISM
#define DANGEROUS_PP
#define V_DANGEROUS_PP
#define PP_STORM
#define PP_ATTACKED

#define PAWN_BLOCK
#define NO_CENTRE_PAWNS
#define CENTRE_PAWN_BONUS
#define NO_PAWNS
Dann Corbit
 


Return to Programming and Technical Discussions

Who is online

Users browsing this forum: Majestic-12 [Bot], MSN [Bot] and 3 guests