Seer

Graham Banks · Post by **Graham Banks** » Wed Nov 18, 2020 6:26 am

connor_mcmonigle wrote: ↑Wed Nov 18, 2020 5:45 am I've released Seer 1.2 here: https://github.com/connormcmonigle/seer ... s/tag/v1.2
It should resolve existing UCI compatibility issues (and hopefully doesn't create any new issues). The improved network adds significant elo over the previous release.

Thanks. Will let you know how it goes.

Graham Banks · Post by **Graham Banks** » Wed Nov 18, 2020 6:34 am

Graham Banks wrote: ↑Wed Nov 18, 2020 6:26 am
connor_mcmonigle wrote: ↑Wed Nov 18, 2020 5:45 am I've released Seer 1.2 here: https://github.com/connormcmonigle/seer ... s/tag/v1.2
It should resolve existing UCI compatibility issues (and hopefully doesn't create any new issues). The improved network adds significant elo over the previous release.
Thanks. Will let you know how it goes.

It works under ChessGUI now.

connor_mcmonigle · Post by **connor_mcmonigle** » Wed Nov 18, 2020 6:41 am

Graham Banks wrote: ↑Wed Nov 18, 2020 6:34 am
Graham Banks wrote: ↑Wed Nov 18, 2020 6:26 am
connor_mcmonigle wrote: ↑Wed Nov 18, 2020 5:45 am I've released Seer 1.2 here: https://github.com/connormcmonigle/seer ... s/tag/v1.2
It should resolve existing UCI compatibility issues (and hopefully doesn't create any new issues). The improved network adds significant elo over the previous release.
Thanks. Will let you know how it goes.
It works under ChessGUI now.

Yay!

Frank Quisinsky · Post by **Frank Quisinsky** » Wed Nov 18, 2020 8:21 am

Hi Graham,

good news for me!

With Seer 1.1 same problems with Shredder GUI.
Seer 1.2 Skylake works fine here on my Intel i9-10900k.

I will put Seer 1.2 Skylake later this day again in my next tourney, start 01. December, 2020.

Thanks to the programmer!

Best
Frank

connor_mcmonigle · Post by **connor_mcmonigle** » Wed Nov 18, 2020 8:43 am

Frank Quisinsky wrote: ↑Wed Nov 18, 2020 8:21 am Hi Graham,

good news for me!

With Seer 1.1 same problems with Shredder GUI.
Seer 1.2 Skylake works fine here on my Intel i9-10900k.

I will put Seer 1.2 Skylake later this day again in my next tourney, start 01. December, 2020.

Thanks to the programmer!

Best
Frank

Thanks you. I'll be excited to see it play

Werner · Post by **Werner** » Wed Nov 18, 2020 10:19 am

Thanks for the new version with +200 !
But I think there is still room for improvements

best wishes

Werner
[Event "40 Züge in 8 min"]
[Site "GreKo, DESKTOP-K5TPRSK"]
[Date "2020.11.18"]
[Round "1.1"]
[White "Seer 1.2NN x64 1CPU"]
[Black "FabChess 1.16 x64 1CPU"]
[ECO "C41"]
[Result "1/2-1/2"]

1. e4 e5 2. Nf3 d6 3. d4 exd4 4. Qxd4 a6 5. Bg5 Nc6 6. Qd2
Nf6 7. Nc3 Be7 8. O-O-O O-O 9. Bc4 b5 10. Bb3 Na5 11. Rhe1
Nxb3+ 12. axb3 Bg4 13. Qf4 {+0.74/22 11s} Qc8 {+0.80/26
17s} 14. h3 {+0.71/22 9s} Bxf3 {+0.58/24 14s} 15. gxf3
{+1.08/20 9s} Re8 {+0.69/25 14s} 16. Rg1 {+1.51/27 2:08m}
Qe6 {+0.69/24 14s} 17. Qg3 {+1.53/23 8s} g6 {+0.52/25 14s}
18. f4 {+1.18/23 8s} b4 {+0.59/25 15s} 19. Bxf6 {+1.13/24
15s} bxc3 {+0.76/27 15s} 20. Bxc3 {+1.50/23 7s} Qxe4
{+0.49/28 17s} 21. Rge1 {+1.48/24 13s} Qf5 {+0.45/28 14s}
22. Qg4 {+1.16/23 7s} Qxg4 {+0.50/28 14s} 23. hxg4
{+1.35/20 7s} Bf8 {+0.45/27 18s} 24. f5 {+1.20/23 18s} Bh6+
{+0.60/23 17s} 25. Kb1 {+0.47/23 10s} Bf4 {+0.45/25 17s}
26. Rh1 {+0.58/24 49s} Rac8 {+0.55/25 17s} 27. Rh3
{+1.92/25 24s} h6 {+0.79/27 17s} 28. Rd4 {+1.72/23 5s} Be5
{+0.55/28 17s} 29. Ra4 {+1.62/21 5s} Bxc3 {+0.72/29 18s}
30. bxc3 {+1.76/21 5s} gxf5 {+0.64/28 14s} 31. gxf5
{+1.76/21 5s} Kg7 {+0.75/28 17s} 32. Kb2 {+1.17/21 6s} Re2
{+0.60/27 17s} 33. f3 {+1.11/21 6s} Re5 {+0.57/28 17s}
34. Rg4+ {+0.96/25 23s} Kf6 {+0.95/28 17s} 35. Rxh6+
{+0.75/21 5s} Kxf5 {+0.59/30 17s} 36. Ra4 {+0.80/22 5s} Ra8
{+0.53/26 17s} 37. Rc4 {+0.85/23 5s} Rc8 {+0.49/27 17s}
38. Rxd6 {+1.25/29 40s} cxd6 {0.00/33 17s} 39. Rxc8
{+0.83/23 3s} Kf4 {0.00/35 17s} 40. b4 {+0.89/23 4s} Kxf3
{0.00/34 17s} 41. c4 {+0.86/28 58s} f5 {0.00/31 11s}
42. Rc6 {+0.76/25 6s} f4 {0.00/33 11s} 43. Rxa6 {+1.60/23
6s} Kg3 {0.00/30 11s} 44. Rxd6 {+2.00/22 6s} f3 {0.00/30
11s} 45. Rg6+ {+1.74/25 9s} Kf4 {0.00/32 11s} 46. Rg1
{+2.11/24 6s} Rg5 {0.00/32 11s} 47. Rf1 {+3.33/25 7s} Rg8
{+0.25/28 11s} 48. Kb3 {+2.92/28 21s} Ke3 {+0.32/27 11s}
49. c5 {+2.80/31 1:30m} Ke2 {+0.36/25 11s} 50. Rh1
{+3.46/23 6s} Rc8 {+0.53/24 11s} 51. Rh7 {+3.19/24 5s} f2
{+0.88/24 15s} 52. Re7+ {+5.06/31 1:11m} Kf1 {+1.28/23 17s}
53. Rf7 {+4.03/27 8s} Ke2 {+1.29/21 12s} 54. Kc4 {+5.64/29
53s} f1=Q {+1.53/20 13s} 55. Rxf1 {+9.74/31 40s} Kxf1
{+3.86/21 14s} 56. b5 {+9.79/24 3s} Ke1 {+5.13/19 12s}
57. b6 {+12.53/27 29s} Rb8 {+14.01/21 10s} 58. Kb5
{+13.44/23 1s} Rd8 {+18.06/24 10s} 59. c6 {+14.12/23 2s}
Rd5+ {+18.51/26 10s} 60. Kb4 {+16.34/27 21s} Rd6 {+18.94/24
10s} 61. Kc5 {+19.88/22 1s} Re6 {+23.86/24 12s} 62. c7
{+20.13/21 1s} Re5+ {+27.07/25 27s} 63. Kd6 {+23.15/24 15s}
Rb5 {+25.19/23 13s} 64. Kc6 {+25.79/23 7s} Rb2 {+30.11/21
10s} 65. c4 {+27.19/23 7s} Rf2 {+31.20/20 9s} 66. c8=Q
{+26.92/20 5s} Rb2 {+146.88/24 17s} 67. b7 {+31.17/21 2s}
Kd2 {+38.86/21 9s} 68. Kd5 {+32.92/21 2s} Kc2 {+M8/32 9s}
69. b8=Q {+256.00/18 1s} Rxb8 {+M7/62 9s} 70. Qxb8
{+256.00/16 0s} Kc3 {+M6/100 9s} 71. Ke4 {+256.00/16 0s}
Kxc4 {+M6/46 9s} 72. Qb6 {+22.44/12 0s} Kc3 {Buch 0s}
73. Qd4+ {+256.00/18 0s} Kc2 {+M4/100 2s} 74. Qb4
{+256.00/18 0s} Kd1 {+M3/100 0s} 75. Kd3 {+256.00/18 0s}
Kc1 {Buch 0s} 76. Qf4+ {+256.00/19 0s} Kb2 {+M3/100 0s}
77. Qb8+ {+256.00/17 0s} Kc1 {+M2/100 0s} 78. Qg8
{+256.00/19 0s} Kb2 {+M4/100 0s} 79. Qa8 {+256.00/17 0s}
Kb3 {+M3/100 0s} 80. Qa5 {+256.00/19 0s} Kb2 {Buch 0s}
81. Qe5+ {+256.00/26 7s} Kb1 {+M3/100 0s} 82. Qe1+
{+256.00/25 7s} Kb2 {+M4/100 0s} 83. Qf2+ {+256.00/23 7s}
Kb3 {+M3/100 0s} 84. Qf8 {+256.00/17 7s} Kb2 {+M4/100 0s}
85. Qh8+ {+256.00/22 7s} Ka3 {+M3/100 0s} 86. Qe5
{+256.00/17 7s} Kb4 {+M4/100 0s} 87. Qb8+ {+256.00/18 7s}
Kc5 {+M5/100 2s} 88. Ke3 {+256.00/18 7s} Kc4 {+19.76/17
11s} 89. Qf4+ {+256.00/17 7s} Kc5 {+19.76/16 11s} 90. Qd4+
{+256.00/18 7s} Kb5 {+M6/37 11s} 91. Qd7+ {+256.00/17 7s}
Kb4 {+M5/100 7s} 92. Qd5 {+256.00/18 7s} Kc3 {+M4/100 0s}
93. Qe5+ {+256.00/19 7s} Kb3 {+M5/100 4s} 94. Qb5+
{+256.00/18 8s} Kc3 {+M4/100 0s} 95. Qa5+ {+256.00/18 8s}
Kc4 {+M6/55 11s} 96. Qa6+ {+256.00/20 8s} Kd5 {+19.76/17
11s} 97. Qf6 {+256.00/20 8s} Kc5 {+M6/42 11s} 98. Kf3
{+256.00/18 8s} Kb5 {+19.72/15 12s} 99. Ke4 {+256.00/18 8s}
Kc5 {+M6/32 11s} 100. Kd3 {+256.00/18 8s} Kd5 {+M5/100 2s}
101. Qe7 {+256.00/24 8s} Kc6 {Buch 0s} 102. Qe6+
{+256.00/18 8s} Kc7 {+M5/100 4s} 103. Kc4 {+256.00/19 8s}
Kb7 {+M4/100 0s} 104. Qe8 {+256.00/18 9s} Kc7 {+M4/100 0s}
105. Qe5+ {+256.00/22 9s} Kb6 {+M5/100 5s} 106. Qe3+
{+256.00/20 9s} Kc7 {+M5/100 4s} 107. Qd2 {+256.00/18 9s}
Kc8 {+M5/100 1s} 108. Qd6 {+256.00/20 9s} Kb7 {Buch 0s}
109. Qd7+ {+256.00/21 9s} Ka8 {+M3/100 0s} 110. Qe8+
{+256.00/21 10s} Kb7 {+M4/100 0s} 111. Kc3 {+256.00/18 10s}
Kc7 {+M6/46 12s} 112. Qe6 {+256.00/19 10s} Kb8 {+M5/100 2s}
113. Kb4 {+256.00/19 10s} Kb7 {+M4/100 0s} 114. Kc5
{+256.00/19 11s} Kc7 {+M3/100 0s} 115. Qd6+ {+256.00/20
11s} Kc8 {+M3/100 0s} 116. Qe7 {+256.00/20 11s} Kb8 {Buch
0s} 117. Qe5+ {+256.00/21 12s} Kb7 {+M3/100 0s} 118. Qe8
{+256.00/21 12s} Kc7 {+M3/100 0s} 119. Qa4 {+256.00/20 13s}
Kc8 {0.00/100 0s} 120. Kc6 {+256.00/24 14s} Kd8 {0.00/100
0s} 121. Kd6 {+256.00/28 8s} Kc8 {Buch 0s} 1/2-1/2

no mate with K+Q - K
50 move rule!!

Daniel Anulliero · Post by **Daniel Anulliero** » Wed Nov 18, 2020 12:34 pm

Curious how nn engines can be very strongs and sometimes can't even mate with KQk , funny

Guenther · Post by **Guenther** » Wed Nov 18, 2020 12:55 pm

Daniel Anulliero wrote: ↑Wed Nov 18, 2020 12:34 pm Curious how nn engines can be very strongs and sometimes can't even mate with KQk , funny 😂

Well, I believe some other programs (even unexpected and not only nn ones) aren't able too, at least without tablebases.

It is really a good idea that testers sometimes check programs with no adjudication, even when this is more time consuming.
If one always tests with adjudication and GUI or engine tablebases he might never see such a problem.

mar · Post by **mar** » Wed Nov 18, 2020 2:11 pm

Guenther wrote: ↑Wed Nov 18, 2020 12:55 pm
Daniel Anulliero wrote: ↑Wed Nov 18, 2020 12:34 pm Curious how nn engines can be very strongs and sometimes can't even mate with KQk , funny
Well, I believe some other programs (even unexpected and not only nn ones) aren't able too, at least without tablebases.

It is really a good idea that testers sometimes check programs with no adjudication, even when this is more time consuming.
If one always tests with adjudication and GUI or engine tablebases he might never see such a problem.

I'm not quite sure, search should find the mate easily.
I made an experiment recently with Cheng, where I fed it with 0.0 evals and with random evals, in both cases the search found the mate in KQk quickly.
(I only tested one random position though)

even in the "seer" position where it fails to deliver a mate in 3 when I simply return 256cp in my eval, a mate is still found in ~0.25 seconds, so KQk is indeed not about eval, but search

JohnWoe · Post by **JohnWoe** » Wed Nov 18, 2020 2:29 pm

Games are adjucated to save resources.
And that's a NN thing since it knows nothing about Rule 50.

I've fixed it with Tiny helper. It gives fast mates now.

Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer

Re: Seer