Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question for Prof. Hyatt

Author: Stuart Cracraft

Date: 10:38:34 02/26/06

Go up one level in this thread


On February 25, 2006 at 19:13:42, Robert Hyatt wrote:

>That looks strange to me.  Here is Crafty on my 750mhz laptop, giving all depth
>outputs:
>
>              depth   time  score   variation (1)
>                1     0.00  -0.60   1. ... Rb7
>                1->   0.00  -0.60   1. ... Rb7
>                2     0.00  -0.42   1. ... Rb7 2. e4
>                2     0.00  -0.48   1. ... c3 2. bxc3 Rxc3
>                2->   0.00  -0.48   1. ... c3 2. bxc3 Rxc3
>                3     0.01  -0.48   1. ... c3 2. bxc3 Rxc3
>                3     0.01  -0.72   1. ... Rb7 2. e4 Rg7
>                3->   0.01  -0.72   1. ... Rb7 2. e4 Rg7
>                4     0.01  -0.40   1. ... Rb7 2. e4 fxe4+ 3. Kxe4 Rd7
>                4     0.01     -1   1. ... c3!!
>                4     0.02  -1.26   1. ... c3 2. bxc3 Rxc3 3. Ra2
>                4->   0.02  -1.26   1. ... c3 2. bxc3 Rxc3 3. Ra2
>                5     0.02  -1.56   1. ... c3 2. bxc3 Rxc3 3. Ra2 Rc2
>                5->   0.02  -1.56   1. ... c3 2. bxc3 Rxc3 3. Ra2 Rc2
>                6     0.03  -1.49   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3
>                6->   0.03  -1.49   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3
>                7     0.04  -1.51   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4
>                7->   0.04  -1.51   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4
>                8     0.06  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>                8->   0.11  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>                9     0.13  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>                9->   0.19  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>               10     0.24  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>               10->   0.52  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>               11     0.58  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>               11->   1.13  -1.58   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3
>               12     1.24     -1   1. ... c3!!
>               12     1.41  -2.10   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3 a3
>               12->   2.70  -2.10   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. h4 Rb3 6. Rxd3 Rxd3
>                                    7. Kxd3 a3
>               13     3.22     -1   1. ... c3!!
>               13     3.55  -2.41   1. ... c3 2. bxc3 Rxc3 3. Rb2 Rxa3
>                                    4. Rb6+ Ke7 5. Rb7+ Ke6 6. Rb6+ Kd5
>                                    7. Rb7 Kd6 8. Rb4 Kc5 9. Rd4 h6 <HT>
>               13->   6.29  -2.41   1. ... c3 2. bxc3 Rxc3 3. Rb2 Rxa3
>                                    4. Rb6+ Ke7 5. Rb7+ Ke6 6. Rb6+ Kd5
>                                    7. Rb7 Kd6 8. Rb4 Kc5 9. Rd4 h6 <HT>
>               14     7.95  -2.44   1. ... c3 2. bxc3 Rxc3 3. Rb2 Rxa3
>                                    4. Rb6+ Kf7 5. Rb7+ Kg6 6. Rb6+ Kh5
>                                    7. Rb7 d2 8. Rxh7+ Kg6 9. Rd7 Ra2 10.
>                                    Rd6+ Kg7 11. h4
>               14->  15.19  -2.44   1. ... c3 2. bxc3 Rxc3 3. Rb2 Rxa3
>                                    4. Rb6+ Kf7 5. Rb7+ Kg6 6. Rb6+ Kh5
>                                    7. Rb7 d2 8. Rxh7+ Kg6 9. Rd7 Ra2 10.
>                                    Rd6+ Kg7 11. h4
>               15    19.49  -2.51   1. ... c3 2. bxc3 Rxc3 3. e4 fxe4+
>                                    4. Kxe4 Rxa3 5. Rb2 Rb3 6. Rxb3 axb3
>                                    7. Kxd3 Kf5 8. Kc3 h6 9. Kxb3 Kxf4
>               15    40.19     -1   1. ... Rxb2!!
>               15    42.07     -3   1. ... Rxb2!!
>               15    47.85  -4.13   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Kc5 6. Rb8 c2 7.
>                                    Rc8+ Kb5 8. Kg3 d2 <HT>
>               15->  54.75  -4.13   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Kc5 6. Rb8 c2 7.
>                                    Rc8+ Kb5 8. Kg3 d2 <HT>
>              time=56.42  mat=0  n=20333227  fh=89%  nps=360K
>              ext-> chk=898427 1rep=18873 mate=193
>              predicted=0  nodes=20333227  evals=11783895  50move=0
>              endgame tablebase-> probes=0  hits=0
>
>
>What looks so wrong is that you like Rxb2 at very early depths.  Here is the
>search output if crafty only considers Rxb2 at the root, no other moves allowed:
>              depth   time  score   variation (1)
>                1     0.00   1.86   1. ... Rxb2 2. Rxb2
>                1->   0.00   1.86   1. ... Rxb2 2. Rxb2
>                2     0.01   1.86   1. ... Rxb2 2. Rxb2
>                2->   0.01   1.86   1. ... Rxb2 2. Rxb2
>                3     0.01     -1   1. ... Rxb2!!
>                3     0.01   0.70   1. ... Rxb2 2. Rxb2 c3
>                3->   0.02   0.70   1. ... Rxb2 2. Rxb2 c3
>                4     0.02   1.00   1. ... Rxb2 2. Rxb2 c3 3. Rb7
>                4->   0.02   1.00   1. ... Rxb2 2. Rxb2 c3 3. Rb7
>                5     0.02   1.00   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kf7
>                                    4. Rb7+ Kg6
>                5->   0.02   1.00   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kf7
>                                    4. Rb7+ Kg6
>                6     0.02   1.18   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kf7
>                                    4. Rb7+ Kg6 5. e4
>                6->   0.02   1.18   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kf7
>                                    4. Rb7+ Kg6 5. e4
>                7     0.02     -1   1. ... Rxb2!!
>                7     0.04   0.62   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kg7
>                                    4. Rb7+ Kg6 5. e4 c2 <HT>
>                7->   0.04   0.62   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kg7
>                                    4. Rb7+ Kg6 5. e4 c2 <HT>
>                8     0.06   0.68   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kg7
>                                    4. Rb7+ Kg6 5. e4 fxe4+ 6. Kxe4 c2
>                8->   0.06   0.68   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Kg7
>                                    4. Rb7+ Kg6 5. e4 fxe4+ 6. Kxe4 c2
>                9     0.13     -1   1. ... Rxb2!!
>                9     0.16   0.18   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rf7 c2 6. Rf6+ Kd7 7.
>                                    Rf7+ Ke6
>                9->   0.18   0.18   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rf7 c2 6. Rf6+ Kd7 7.
>                                    Rf7+ Ke6
>               10     0.26  -0.01   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Ke7
>               10->   0.27  -0.01   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Ke7
>               11     0.38  -0.01   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Ke7
>               11->   0.39  -0.01   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Ke7
>               12     0.85  -0.01   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Ke7
>               12->   0.86  -0.01   1. ... Rxb2 2. Rxb2 c3 3. Rb6+ Ke7
>                                    4. Rb7+ Kd6 5. Rb6+ Ke7
>              time=0.88  mat=0  n=242418  fh=95%  nps=275K
>              ext-> chk=12385 1rep=324 mate=0
>
>
>
>So early on, Rxb2 looks very bad for black (good for white) until it thinks it
>finds a repetition...
>
>So the question has to be, why does your program like Rxb2 so early???  It drops
>a rook outright.

Bad bugs in the pawn evaluation logic caused it.
The bugs were the failure to separate attack-based issues in
pawn structure evaluation from the pawn hash and also simply
a wrong coefficient for a term altogether.

Now that the former are separate (attacks on weak pawns,
attacks/defenses/occupation of the square in front of passers, etc),
the search is better although I do not particularly like having to
search to 12 ply to find the solution.

Here's my latest (more normal) search.

.sd 99
maxdepth = 99 maxtime = 9999999 timed = 0
.ts
position file? [wac.epd] wac.epd
# of test positions to test? 1
maxtime = 9999999
Interrupt current ply and return move at timeout
Testsuite: Tests/wac.epd 1 positions
*** Problem   Solution(s): Rxb2 (bm)
-- ** -- ** -- ** -- **
** -- ** -- ** -- ** BP
-- ** -- ** -- BK -- **
** -- ** -- ** BP ** --
BP ** BP ** -- WP -- **
WP BR ** BP WP WK ** --
-- WP -- WR -- ** -- WP
** -- ** -- ** -- ** --
mv 1 stage 0, black to move, computer plays black
hash=359a65d864d93847
pawnhash=3e87fdd09deedce8
0 0 0 0 0 0
Alpha=-593 Beta=207 Maxdepth=99 MaxTime=9999999 xboard=0
Itr/Max Mv   Score   Time Nodes PV
 1/ 2  f6e6 124  0.02       20 f6e6
 2/ 3  b3b7 86  0.02       51 b3b7
 3/ 4  h7h5 -208  0.02      280 h7h5 h2h3
 4/ 7  f6e6 -151  0.03      737 f6e6 e3e4
 5/10  f6e6 -19  0.04     1851 f6e6 d2g2 e6d6
 6/13  c4c3 297  0.07     6754 c4c3 b2c3 b3c3 d2b2 c3a3
 7/16  c4c3 99  0.12    14647 c4c3 b2c3 b3c3 d2b2 c3a3 b2b7 d3d2
 8/16  c4c3 175  0.20    27242 c4c3 b2c3 b3c3 d2b2 c3a3 b2b7 f6g6 b7b6 g6h5
 9/16  c4c3 245  0.35    49376 c4c3 b2c3 b3c3 e3e4 c3a3 e4f5 f6f5 f3e3
10/26  c4c3 141  3.37   475736 c4c3 b2c3 b3c3 e3e4 c3a3 f3e3 f5e4 e3e4 a3c3 d2d3
c3c2
11/26  c4c3 224  4.83   691883 c4c3 b2c3 b3c3 e3e4 c3a3 f3e3 a3a1 e4f5 f6f5 d2d3
a1a2 d3d5
12/30  c4c3 300 27.89  3918835 c4c3 b2c3 b3c3 d2b2 c3b3 b2a2 h7h5 h2h3 f6e6 a2g2
13/34  b3b2 491 258.84 35767718 b3b2 d2b2 c4c3 b2b6 f6g7 b6b1 c3c2 b1g1 g7h6
g1a1 d3d2 f3e2 c2c1R a1
c1
14/36  b3b2 491 502.64 70319840 b3b2 d2b2 c4c3 b2b6 f6g7 b6b7 g7f6 b7b6 f6g7
b6b7 g7f6 b7b6
15/...



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.