Telepath mistake: test position

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

CRoberson
Posts: 2094
Joined: Mon Mar 13, 2006 2:31 am
Location: North Carolina, USA

Telepath mistake: test position

Post by CRoberson »

In the following, at G/120 + 10, Telepath played Qxg4 (p) instead of Qxh4 (knight). This is a mistake. I thought that taking the knight was
obvious, but evidently not. Turns out Telepath liked Qxh4 for a while then went with Qxg4. Under analysis mode it does change back to Qxh4
but after the time to move threshold passed. I spent time trying to figure out how it could be so dumb, then tried Rybka and it does the same thing,
but it would have changed back to Qxh4 in time.

So, how many engines toggle between Qxh4 and Qxg4 and look at the times. It is interesting to see that an engine would make a mistake
in a game if the time control were G/5 or G/10.

FYI: the game happened on ICC yesterday between Crafty and Telepath at G/120 + 10.

[d] r3kbnr/1p1b1ppp/p1np4/3Bp2q/4P1PN/1Q2B3/PPP2P1P/RN2K2R b KQkq g3 0 11
Top
kgburcham
Posts: 2016
Joined: Sun Feb 17, 2008 4:19 pm

DF1 & DF11

Post by kgburcham »

[d] r3kbnr/1p1b1ppp/p1np4/3Bp2q/4P1PN/1Q2B3/PPP2P1P/RN2K2R b KQkq - 0 1

Deep Fritz 1:

1...Qxg4 2.Bxf7+ Kd8 3.Nd2 Nh6 4.Nhf3 Nxf7 5.Qxf7 Ne7 6.Qb3 Bc6 7.Rg1
+/= (0.62) Depth: 11/33 00:00:01 12254kN
1...Qxg4 2.Bxf7+ Kd8 3.Nd2 Nh6 4.Nhf3 Nxf7 5.Qxf7 Kc7 6.Qb3 Kb8 7.Rg1
+/= (0.59) Depth: 12/37 00:00:02 26992kN
1...Qxg4 2.Bxf7+ Kd8 3.Nd2 Nh6 4.Bxh6 Nd4 5.Qd3 Rc8 6.Rc1 Qxh4 7.Be3
+/= (0.66) Depth: 13/39 00:00:06 65563kN
1...Qxg4 2.Bxf7+ Kd8 3.Nd2 Nh6 4.Bxh6 Nd4 5.Qd3 Qxh4 6.Be3 Qf6 7.Bc4
+/= (0.69) Depth: 14/40 00:00:15 172mN
1...Qxg4
+/- (1.00) Depth: 15/45 00:00:39 415mN
1...Qxg4 2.Bxf7+ Kd8 3.Bxg8 Qxh4 4.Bd5 Be7 5.Nd2 Rb8 6.0-0-0 Qf6 7.Rhg1
+/- (1.06) Depth: 15/45 00:01:00 638mN
1...Qxh4
+/- (1.03) Depth: 15/45 00:01:15 801mN
1...Qxh4
+/- (0.72) Depth: 16/51 00:03:07 1985mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8
+/= (0.66) Depth: 16/51 00:03:54 2495mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8
+/- (0.91) Depth: 17/47 00:06:05 3862mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd8 7.Qh8
+/- (0.72) Depth: 18/51 00:16:39 10623mN



Deep Fritz 11:

1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Rh8 5.Nd5+ Kd8 6.Qb6+ Ke8 7.Nc7+ Kf7 8.Qb3+ Kg6 9.Nxa8 Qxg4 10.f3 Qh4+ 11.Bf2 Qf6 12.Qxb7 Qf7 13.Rg1+
+/- (0.78) Depth: 17/38 00:00:02 23446kN
1...Qxg4 2.Bxf7+ Kd8 3.Nc3 Nd4 4.Qxb7 Rc8 5.Nf5 Bxf5 6.exf5 Nxc2+ 7.Kd2 Nd4 8.Bxd4 Qxd4+ 9.Ke1 Nf6 10.Be6 Rxc3 11.bxc3 Qxc3+ 12.Ke2 Qc2+
+/= (0.62) Depth: 17/42 00:00:05 54612kN
1...Qxg4 2.Bxf7+ Kd8 3.Nc3 Nd4 4.Qxb7 Rc8 5.Nf5 Bxf5 6.exf5 Nxc2+ 7.Kd2 Nd4 8.Bxd4 Qxd4+ 9.Ke1 Nf6 10.Be6 Rxc3 11.bxc3 Qxc3+ 12.Ke2 Qc2+ 13.Kf3 Qd3+ 14.Kg2 Qe4+
+/- (0.82) Depth: 18/41 00:00:08 86827kN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8 Bxg4 8.Rg1 Nd4 9.Kd2 Nf3+ 10.Kc3 Nxg1 11.Rxg1 Be2 12.Rxg7+
+/= (0.41) Depth: 18/41 00:00:09 97875kN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8 Bxg4 8.Rg1 Nd4 9.Kd2 Nf3+ 10.Kc3 Nxg1 11.Rxg1 Be2 12.Rxg7+ Bxg7 13.Qxg7+
+/= (0.41) Depth: 19/41 00:00:11 116mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8 Bxg4 8.Rg1 Nd4 9.Kd2 Nf3+ 10.Kc3 Nxg1 11.Rxg1 Kc8 12.Kb3 h5
+/= (0.37) Depth: 20/44 00:00:17 181mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8 Bxg4 8.Rg1 Nd4 9.Kd2 Nf3+ 10.Kc3 Nxg1 11.Rxg1 Kc8 12.Kb3
+/= (0.43) Depth: 21/45 00:00:27 291mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.Qxg8 Be6 6.Nd5+ Kd7 7.Qh8 Bxg4 8.Rg1 Nd4 9.Kd2 Nf3+ 10.Kc3 Nxg1 11.Rxg1 Kc8 12.Kb3 Qh5
+/= (0.41) Depth: 22/47 00:00:55 587mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.0-0-0 Rh8 6.Nd5+ Kf7 7.Nc7+ Ke7 8.f4 Rb8 9.fxe5 Nxe5 10.h3 Kd8 11.Nxa6 Kc8 12.Nxb8
= (0.25) Depth: 23/52 00:03:01 1909mN
1...Qxh4 2.Bxf7+ Ke7 3.Bxg8 Rxg8 4.Nc3 Re8 5.0-0-0 Rh8 6.Nd5+ Kf7 7.Nc7+ Be6 8.Qxb7 Nd8 9.Qxa6 Bxg4 10.f3 Bd7 11.Nxe8 Bxe8 12.Rhg1 Ne6
+/= (0.30) Depth: 24/50 00:05:00 3161mN
Top
User avatar
michiguel
Posts: 6401
Joined: Thu Mar 09, 2006 8:30 pm
Location: Chicago, Illinois, USA

Re: Telepath mistake: test position

Post by michiguel »

CRoberson wrote:In the following, at G/120 + 10, Telepath played Qxg4 (p) instead of Qxh4 (knight). This is a mistake. I thought that taking the knight was
obvious, but evidently not. Turns out Telepath liked Qxh4 for a while then went with Qxg4. Under analysis mode it does change back to Qxh4
but after the time to move threshold passed. I spent time trying to figure out how it could be so dumb, then tried Rybka and it does the same thing,
but it would have changed back to Qxh4 in time.

So, how many engines toggle between Qxh4 and Qxg4 and look at the times. It is interesting to see that an engine would make a mistake
in a game if the time control were G/5 or G/10.

FYI: the game happened on ICC yesterday between Crafty and Telepath at G/120 + 10.

[d] r3kbnr/1p1b1ppp/p1np4/3Bp2q/4P1PN/1Q2B3/PPP2P1P/RN2K2R b KQkq g3 0 11

Gaviota sticks to Qxh4 from ply 1. However, at higher plies the difference between one move and the other is not so high. I do not think that Qxg4 is much worse than Qxh4.

Miguel

Code: Select all

analyze
********* Starts iterative deepening, thread = 0
set timer to infinite
        57   1       0.0    -0.37  Qxh4 7.Bxf7+ Kd8 8.Bxg8 Qxg4
       289   2       0.0    -0.47  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Qxg4
      1561   3       0.0    -0.47  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Qxg4
      3508   4       0.0      :-(  Qxh4
      7782   4       0.1    -1.08  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8
                                   Bxg4
     16598   5       0.1    -1.08  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8
                                   Bxg4
     50721   6       0.4      :-)  Qxh4
     74426   6       0.6    -0.62  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Bxg4
     94268   6:      0.7    -0.62  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Bxg4
    176354   7       1.2    -0.62  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Bxg4
    266676   7:      1.5    -0.62  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Bxg4
    464470   8       1.9    -0.95  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Nd4 11.Bxd4 exd4
   1258379   8:      3.6    -0.95  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Nd4 11.Bxd4 exd4
   2248282   9       5.7    -0.83  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Bxg4 11.Nc3 Bf3
   2833430   9:      7.0    -0.83  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Bxg4 11.Nc3 Bf3
   4184734  10       9.8    -0.92  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Qxg4 11.Qxh7 Qg2 12.Qh4+ Kd7
   6217247  10:     14.2    -0.92  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Qxg4 11.Qxh7 Qg2 12.Qh4+ Kd7
  12264029  11      26.7    -0.95  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Nd4 11.Bxd4 exd4 12.Nd2 Rc8
                                   13.O-O-O Bxa2
  21204646  11:     45.3    -0.95  Qxh4 7.Bxf7+ Ke7 8.Bxg8 Rxg8 9.Qxg8 Be6
                                   10.Qh8 Nd4 11.Bxd4 exd4 12.Nd2 Rc8
                                   13.O-O-O Bxa2
 101577618  12     206.0    -0.83  Qxh4 7.Bxf7+ Kd8 8.Bxg8 Qxg4 9.Nd2 b5
                                   10.h3 Qg2 11.Ke2 Nd4+ 12.Bxd4 exd4
                                   13.Rag1 d3+ 14.cxd3
 131400673  12:    268.4    -0.83  Qxh4 7.Bxf7+ Kd8 8.Bxg8 Qxg4 9.Nd2 b5
                                   10.h3 Qg2 11.Ke2 Nd4+ 12.Bxd4 exd4
                                   13.Rag1 d3+ 14.cxd3
 274783619  13     546.4    -0.88  Qxh4 7.Bxf7+ Kd8 8.Bxg8 Qxg4 9.Nd2 b5
                                   10.h3 Qh5 11.Bd5 Be7 12.a4 Rb8 13.axb5
                                   axb5
 438176919  13:    869.2    -0.88  Qxh4 7.Bxf7+ Kd8 8.Bxg8 Qxg4 9.Nd2 b5
                                   10.h3 Qh5 11.Bd5 Be7 12.a4 Rb8 13.axb5
                                   axb5
1045859877  14    2133.1    -1.01  Qxh4 7.Bxf7+ Kd8 8.Bxg8 Qxg4 9.Nd2 b5
                                   10.h3 Qh5 11.Bd5 Rb8 12.Qa3 Nd4 13.Bxd4
                                   exd4 14.h4
After Qxg4

Code: Select all

********* Starts iterative deepening, thread = 0
set timer to infinite
        17   1       0.0    -1.86  7.Bxc6 Bxc6
       105   1       0.0    +1.26  7.Bxf7+ Ke7 8.Nf5+ Bxf5 9.exf5
       505   2       0.0    +1.19  7.Bxf7+ Kd8 8.Bxg8 Qxh4 9.Qxb7
      1225   3       0.0    +1.19  7.Bxf7+ Kd8 8.Bxg8 Qxh4 9.Qxb7
      7352   4       0.0    +1.31  7.Bxf7+ Kd8 8.Bxg8 Qxh4 9.Nc3
     35428   5       0.1    +0.96  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Qxb7
    100131   6       0.3    +0.96  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Qxb7
    166776   6:      0.5    +0.96  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Qxb7
    414131   7       1.1    +1.20  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Rg1 Qf5 11.Qxb7
    609103   7:      1.6    +1.20  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Rg1 Qf5 11.Qxb7
   2216425   8       5.2    +1.33  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Rg1 Qf5 11.Nxe4 Qxe4 12.Qxb7
   2568231   8:      6.0    +1.33  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4
                                   10.Rg1 Qf5 11.Nxe4 Qxe4 12.Qxb7
   8205779   9      18.9    +1.56  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4 10.h3
                                   Qf5 11.Nh4 Qf6 12.Nxe4 Qxh4 13.Qxb7
                                   Qxe4 14.Qxa8+ Bc8
   9118054   9:     20.9    +1.56  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4 10.h3
                                   Qf5 11.Nh4 Qf6 12.Nxe4 Qxh4 13.Qxb7
                                   Qxe4 14.Qxa8+ Bc8
  21229687  10      46.3      :-(  7.Bxf7+
  27628843  10      60.0      :-(  
  33971628  10      73.9    +1.56  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4 10.h3
                                   Qf5 11.Nh4 Qf6 12.Nxe4 Qxh4 13.Qxb7
                                   Qxe4 14.Qxa8+ Bc8
  34061190  10:     74.1    +1.56  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4 10.h3
                                   Qf5 11.Nh4 Qf6 12.Nxe4 Qxh4 13.Qxb7
                                   Qxe4 14.Qxa8+ Bc8
  52517597  11     112.1      :-(  7.Bxf7+
  74226609  11     158.0      :-(  
 102593649  11     218.0    +1.08  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4 10.h3
                                   Nxd2 11.Qb6+ Ke7 12.hxg4 Nxf3+ 13.Kf1
                                   Kxf7 14.Qb3+ Kg6 15.Qxb7
 133363071  11:    280.3    +1.08  7.Bxf7+ Kd8 8.Nf3 Nf6 9.Nbd2 Nxe4 10.h3
                                   Nxd2 11.Qb6+ Ke7 12.hxg4 Nxf3+ 13.Kf1
                                   Kxf7 14.Qb3+ Kg6 15.Qxb7
Top
CRoberson
Posts: 2094
Joined: Mon Mar 13, 2006 2:31 am
Location: North Carolina, USA

Re: Telepath mistake: test position

Post by CRoberson »

Hi Miguel,

A deceiving position it is. Let it run deep. Qxh4 approaches 0 with each iteration. Qxg4 approaches a score of +1.5 to +2.nn for white.

Let it run to 19 ply or so.

Better yet eval the leaf nodes in these two lines

... Qxg4 Bxf7 Kd8
... Qxh4 Bxf7 Ke7
Top
bob
Posts: 20943
Joined: Mon Feb 27, 2006 7:30 pm
Location: Birmingham, AL

Re: Telepath mistake: test position

Post by bob »

CRoberson wrote:In the following, at G/120 + 10, Telepath played Qxg4 (p) instead of Qxh4 (knight). This is a mistake. I thought that taking the knight was
obvious, but evidently not. Turns out Telepath liked Qxh4 for a while then went with Qxg4. Under analysis mode it does change back to Qxh4
but after the time to move threshold passed. I spent time trying to figure out how it could be so dumb, then tried Rybka and it does the same thing,
but it would have changed back to Qxh4 in time.

So, how many engines toggle between Qxh4 and Qxg4 and look at the times. It is interesting to see that an engine would make a mistake
in a game if the time control were G/5 or G/10.

FYI: the game happened on ICC yesterday between Crafty and Telepath at G/120 + 10.

[d] r3kbnr/1p1b1ppp/p1np4/3Bp2q/4P1PN/1Q2B3/PPP2P1P/RN2K2R b KQkq g3 0 11
First, I looked at the log. Crafty played g4, expecting Qxh4:

Code: Select all

               20     4:00   1.31   11. g4 Qxh4 12. Bxf7+ Kd8 13. Bxg8
                                    Qxg4 14. Nd2 b5 15. Bb6+ Kc8 16. f3
                                    Qh5 17. O-O-O Rb8 18. Bf7 Qh4 19. Be3
                                    Rb7 20. Qa3 a5 21. Bd5 b4
              time=4:00  mat=0  n=4977121246  fh=91%  nps=20.7M
              extensions=178.5M qchecks=178.7M reduced=475.3M pruned=2150.0M
              predicted=7  evals=2207.5M  50move=9  EGTBprobes=0  hits=0
I then ran the position on the same machine, starting from scratch:

Code: Select all

              18    29.54     +1   1. ... Qxg4?                          
               18    46.34   2.01   1. ... Qxg4 2. Bxf7+ Ke7 3. Bxg8 Rxg8 
                                    4. Nc3 Qxh4 5. O-O-O Rh8 6. Nd5+ Kf7
                                    7. Nc7+ Kf6 8. Nxa8 Qxe4 9. Nc7 Na5
                                    10. Nd5+ Kg6 11. Qb6 Qa4 12. Rhg1+
                                    Bg4
               18     1:18   1.26   1. ... Qxh4 2. Bxf7+ Kd8 3. Bxg8 Qxg4 
                                    4. Nd2 b5 5. f3 Qh4+ 6. Bf2 Qf4 7.
                                    O-O-O Nd4 8. Bxd4 exd4 9. Bd5 Ra7 10.
                                    Qb4 d3 11. Kb1 dxc2+ 12. Kxc2
               18->   1:19   1.26   1. ... Qxh4 2. Bxf7+ Kd8 3. Bxg8 Qxg4 
                                    4. Nd2 b5 5. f3 Qh4+ 6. Bf2 Qf4 7.
                                    O-O-O Nd4 8. Bxd4 exd4 9. Bd5 Ra7 10.
                                    Qb4 d3 11. Kb1 dxc2+ 12. Kxc2 (s=2)
               19     1:37   1.32   1. ... Qxh4 2. Bxf7+ Kd8 3. Bxg8 Qxg4 
                                    4. Nd2 b5 5. f3 Qh4+ 6. Bf2 Qf4 7.
                                    O-O-O Nd4 8. Bxd4 exd4 9. Bd5 Ra7 10.
                                    Qb4 d3 11. Qd4 dxc2 12. Kxc2
               19->   1:44   1.32   1. ... Qxh4 2. Bxf7+ Kd8 3. Bxg8 Qxg4 
                                    4. Nd2 b5 5. f3 Qh4+ 6. Bf2 Qf4 7.
                                    O-O-O Nd4 8. Bxd4 exd4 9. Bd5 Ra7 10.
                                    Qb4 d3 11. Qd4 dxc2 12. Kxc2 (s=2)
               20     2:19   1.29   1. ... Qxh4 2. Bxf7+ Kd8 3. Bxg8 Qxg4 
                                    4. Nd2 b5 5. f3 Qh4+ 6. Bf2 Qf4 7.
                                    Bd5 Ke8 8. O-O-O Rc8 9. Be3 Nd4 10.
                                    Bxd4 exd4 11. Qa3 Qe3 12. Qxa6
               20->   2:27   1.29   1. ... Qxh4 2. Bxf7+ Kd8 3. Bxg8 Qxg4 
                                    4. Nd2 b5 5. f3 Qh4+ 6. Bf2 Qf4 7.
                                    Bd5 Ke8 8. O-O-O Rc8 9. Be3 Nd4 10.
                                    Bxd4 exd4 11. Qa3 Qe3 12. Qxa6
Does take a bit of time to see that Qg4 is bad. About 30 seconds at 20M nodes per sec on the machine used yesterday.
Top

Return to “Computer Chess Club: General Topics”