Thinker 5.3B inert-Rybka 232a, 35-15

[Mike S.]

3m+3s, D945 3.4 GHz (P4-like)
engines on single core,
ponder off, 128 MB hash each
Top-25 openings set by S.Canbaz

3/4- and some 5-men tablebases
128 MB tablebase cache,
tablebases on flash memory
(tbs. access by Rybka only)

Code: Select all

1   Rybka 2.3.2a 32-bit    +23/-3/=24 70.00%   35.0/50
2   Thinker 5.3B-32 inert  +3/-23/=24 30.00%   15.0/50

The intermediate result after 20 games was even worse, 15.5-4.5 = 22.5% for Thinker. I want to point to the high number of draws though: Thinker did not lose the majority of games (27/50). Rybka 232a is of course a mighty opponent. I have had a promising match result from Thinker 5.2e passive against Shredder 11 SE:

http://talkchess.com/forum/viewtopic.ph ... ew=threads

In that match, Thinker had won 54.5-45.5 (it was only 1m+1s though); that is why I decided to try Rybka 232a as an opponent for the new version. But Thinker's performance in this match indicates about 150 Elo less. Here is an overview of some ratings:

Code: Select all

Engine (single 32-bit)   CCRL 40/4m   CEGT 40/4m   Average
----------------------------------------------------------------------
Rybka 3 (si.32)                 3127              3039          3083
Rybka 2.3.2a                    3020              2960          2990
Shredder 11 (-SE)             2963              2858          2911
Thinker 5.2I or -M pas.      2886              2824          2855
Thinker 5.1c passive          2842              2789         2816

(sorry for the broken formatting, it's a nightmare)

I do not know if my result against R232a is typical for Thinker 5.3B inert, but if it is, it would get a rating somewhere between the passive versions 5.1 and 5.2, only. - Nevertheless, I believe that Thinker is a very good and interesting engine, which deserves great attention.

[Event "Thinkertest 3m+3s"]
[Site "COMPUTER-11"]
[Date "2008.11.11"]
[Round "41"]
[White "Thinker 5.3B-32 inert"]
[Black "Rybka 2.3.2a 32-bit"]
[Result "1-0"]
[ECO "E15"]
[PlyCount "107"]

1. d4 Nf6 2. c4 e6 3. Nf3 b6 4. g3 Ba6 5. b3 Bb4+ 6. Bd2 Be7 7. Nc3 O-O 8. e4
Bb7 9. Bd3 d5 10. cxd5 exd5 11. e5 Ne4 12. O-O Qd7 13. Re1 Nxd2 14. Qxd2 Nc6
15. a3 Na5 16. Rab1 Bxa3 17. Nb5 Be7 18. e6 fxe6 19. Ne5 Qd8 20. Qc2 Rf6 21.
Bxh7+ Kh8 22. Nxc7 Rc8 23. Ng6+ Kxh7 24. Nxe7+ g6 25. Nxc8 Qxc8 26. Rbc1 Nc6
27. Nb5 Qf8 28. Qd2 a6 29. Nc3 Qb4 30. Red1 Rf3 31. h4 a5 32. Kg2 Rf5 33. g4
Rf7 34. Qg5 Qf8 35. Rc2 e5 36. Nb5 exd4 37. Re2 Ba6 38. h5 Rf6 39. Qxd5 Qg8 40.
Qxg8+ Kxg8 41. Re8+ Kf7 42. g5 Bxb5 43. gxf6 gxh5 44. Rh8 Kxf6 45. Kf3 a4 46.
bxa4 Bxa4 47. Rd2 Ne5+ 48. Kg3 Nc6 49. Rxh5 Kg6 50. Kg4 Kf7 51. f4 Ke8 52. Rh7
Bb3 53. Rb2 Be6+ 54. f5 1-0

I had the impression that Thinker's time consumption is more flexible than Rybka's. While Rybka usually kept a large amount of remaining time as a reserve (basically wasting chances to search deeper), Thinker used more time. For example, in the following game after the shock 20.Bxh6! (which started a winning attack), Thinker invested 1:26 in a 3+3 game for the reply. It didn't save this particular game, but in general I think this time consumption behaviour is good.

[Event "Thinkertest 3m+3s"]
[Site "COMPUTER-11"]
[Date "2008.11.11"]
[Round "18"]
[White "Rybka 2.3.2a 32-bit"]
[Black "Thinker 5.3B-32 inert"]
[Result "1-0"]
[ECO "C02"]
[PlyCount "89"]

1. d4 e6 2. e4 d5 3. e5 c5 4. c3 Nc6 5. Nf3 Nge7 6. Bd3 Bd7 7. O-O Ng6 8. Qe2
Be7 9. dxc5 Bxc5 10. Na3 O-O 11. Nb5 a6 12. Nbd4 Bxd4 13. cxd4 Qb6 14. Be3 Nb4
15. a4 Rfc8 16. Bb1 h6 17. a5 Qb5 18. Qd2 Rc6 19. Ra3 b6 20. Bxh6 Rac8 21. Bxg6
fxg6 22. Qg5 Be8 23. Bxg7 R8c7 24. Bh6 Rc2 25. axb6 Qd7 26. Qf6 Qe7 27. bxc7
Rxc7 28. Rc1 Qxf6 29. exf6 Rxc1+ 30. Bxc1 Kf7 31. Bg5 Nc6 32. Rxa6 Kf8 33. Bh6+
Kf7 34. Bg7 g5 35. Rxc6 Bxc6 36. Ne5+ Kg8 37. Nxc6 g4 38. Ne5 g3 39. fxg3 Kh7
40. b4 Kg8 41. Bh6 Kh7 42. f7 Kxh6 43. f8=Q+ Kg5 44. Qf4+ Kh5 45. Qh4# 1-0

The following position arose after the 111th move of a long draw:

[d]1r6/2rb4/p2p1qk1/n1pPp1p1/PpP1Pp1p/1P1N1P1P/1R1QB1P1/R5K1 w - - 0 112

Thinker was playing "antichess" here, in the style of Pablo Restrepo! The material balance was not changed anymore after the last capturing move, 28...Qxf4. I don't know how long this game would have lasted, but I had configured early draw and max. calculated plies 200.

[Event "Thinkertest 3m+3s"]
[Site "COMPUTER-11"]
[Date "2008.11.11"]
[Round "25"]
[White "Thinker 5.3B-32 inert"]
[Black "Rybka 2.3.2a 32-bit"]
[Result "1/2-1/2"]
[ECO "C97"]
[PlyCount "222"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 d6 8. c3
O-O 9. h3 Na5 10. Bc2 c5 11. d4 Qc7 12. d5 Bd7 13. b3 Rfe8 14. Be3 g6 15. Qc1
Nh5 16. Bh6 Bf6 17. Nbd2 Nf4 18. Nf1 Rab8 19. Ng3 Bg7 20. Bxg7 Kxg7 21. Ne2 Nh5
22. Qd2 Qd8 23. Rac1 Qf6 24. Qd3 Rb7 25. Qe3 h6 26. Qd3 Rc8 27. Red1 Nf4 28.
Nxf4 Qxf4 29. Qe2 Rbc7 30. Rd2 Rb8 31. Rb1 h5 32. Rd3 Qf6 33. Re3 h4 34. Bd1
Rcc8 35. Qd2 Rh8 36. Be2 Rbc8 37. c4 Qd8 38. Bd3 Rb8 39. Ree1 Qb6 40. Red1 Rh5
41. a4 b4 42. Re1 Qd8 43. Re3 Qf6 44. Be2 Qd8 45. Qb2 Qe7 46. Rf1 Re8 47. Ne1
Rg5 48. Nd3 Kg8 49. Kh2 Qd8 50. Ne1 Qe7 51. Nf3 Rh5 52. Rc1 Kg7 53. Qc2 Rb8 54.
Kg1 Rf8 55. Re1 Rhh8 56. Rb1 Rh5 57. Ne1 Rg5 58. Rc1 Bc8 59. Nf3 Rh5 60. Qd1
Bd7 61. Ra1 Rc8 62. Ne1 Rhh8 63. Qc2 Qf6 64. Rb1 Rc7 65. Nd3 Qg5 66. Qb2 Kh7
67. Rd1 Rf8 68. Rf1 Re8 69. Ne1 Kg8 70. Kh1 Qf6 71. Nd3 Qg5 72. Rd1 Qh6 73. Ne1
Rf8 74. Kg1 Qg5 75. Nf3 Qf4 76. Ra1 Rb7 77. Ne1 Qg5 78. Rc1 Kg7 79. Rd1 Qd8 80.
Ra1 Qf6 81. Nd3 Re8 82. Rb1 Rc7 83. Rf1 Re7 84. Re1 Rb7 85. Qc2 Re8 86. Rb1 Qg5
87. Kh1 Rb6 88. Rd1 Rh8 89. Rf1 Re8 90. Qb2 Qf6 91. f3 Rbb8 92. Bd1 Re7 93. Kg1
Rh8 94. Re2 Ree8 95. Ref2 g5 96. Qd2 Ra8 97. Re2 Rac8 98. Bc2 Rb8 99. Qe3 Qe7
100. Rd1 Qf6 101. Qd2 Qe7 102. Ra1 Qf6 103. Ree1 Qg6 104. Qe3 Rhe8 105. Rf1 f5
106. Rab1 f4 107. Qd2 Rh8 108. Rb2 Qf6 109. Ra1 Kg6 110. Bd1 Rhc8 111. Be2 Rc7
1/2-1/2

[swami]

That was a twist! After reading the subject title I jumped to conclusion. (Thinker 35 - Rybka 15)

For a moment I thought "Whoah, Is Thinker really this stronger than Rybka or was there a bug in the set up" but then you spoiled it for me.

The Results table in the message says otherwise, opposite of what was displayed in the thread title. (Thinker 15 - Rybka 35)

[Ovyron]

Thinker 5.3B inert-Rybka 232a, 35-15

By the title I thought Thinker did destroy Rybka.

EDIT - Swami beat me to it. Anyway, I hope Thinker becomes Top 2 freeware after the dust settles.

[Mike S.]

Oops, yes, sorry I guess I need to apologize. The result was Rybka 35 - Thinker 15. I wrote it in the wrong order. Sorry!

[S.Taylor]

Thinker 5.3B inert-Rybka 232a, 35-15

,
By the title I thought Thinker did destroy Rybka.

EDIT - Swami beat me to it. Anyway, I hope Thinker becomes Top 2 freeware after the dust settles.

Beat ME to it too!
Thinker might have alot going for it, as it is already close to top range, and if that is together with other improvements, it might be a sign of promise, even for closing the gap.
But, if this score HAD been the other way round, as mistakenly hinted at in the heading of this thread, then it MIGHT have meant that it is even a little stronger than Rybka 3.
At the momment, it's still "mights". (re. being top in the world, and in history, but a great program is already on the ground)

[swami]

Oops, yes, sorry I guess I need to apologize. The result was Rybka 35 - Thinker 15. I wrote it in the wrong order. Sorry!

Nah, No Problem. You just gave us a little surprise with this... We were happy about Thinker for a moment.

[S.Taylor]

Oops, yes, sorry I guess I need to apologize. The result was Rybka 35 - Thinker 15. I wrote it in the wrong order. Sorry!

Nah, No Problem. You just gave us a little surprise with this... We were happy about Thinker for a moment.

Not only that. Not only the actual rating shows how strong it is. But, IF there is any program which is very high elo, but still not in the top 5, but almost, but sometimes finds moves which are best, that no other computer finds, Then such an engine is indespensible, as you can't be sure that even Rybka 3 will come up with the right move that such a program WOULD come up with. So you need both.
But, if all the best moves are found by one program, that makes life easier, as then you can afford to not need anything besides that program, if you are trying to find out what is the best move.
Every programmer should try either 1).to make theirs THE best in every move OR, 2). Try to stop any other program from being quite that.

[S.Taylor]

Oops, yes, sorry I guess I need to apologize. The result was Rybka 35 - Thinker 15. I wrote it in the wrong order. Sorry!

Nah, No Problem. You just gave us a little surprise with this... We were happy about Thinker for a moment.

Not only that. Not only the actual rating shows how strong it is. But, IF there is any program which is very high elo, but still not in the top 5, but almost, but sometimes finds moves which are best, that no other computer finds, Then such an engine is indespensible, as you can't be sure that even Rybka 3 will come up with the right move that such a program WOULD come up with. So you need both.
But, if all the best moves are found by one program, that makes life easier, as then you can afford to not need anything besides that program, if you are trying to find out what is the best move.
Every programmer should try either 1).to make theirs THE best in every move OR, 2). Try to stop any other program from being quite that.