preventing 3-fold repetition

[flok]

Hi,

In the tourney of January (http://talkchess.com/forum/viewtopic.ph ... ight=embla), it was mentioned that embla lost by a 3-fold repetition.

To make sure that won't happen again, I experimented a bit with it.
The 3-fold rep. was with "if position occured before, use score 0".
Now I tried with "skip the move" and "score = -min_mate_value". These two experiments both played far worse (-300 elo) than the score=0 version.

What would be the best to do?

[hgm]

I thought 3-fold repetition always was a draw in Chess. Only in Shogi and Xiangqi it can be losing. So how could you lose by 3-fold repetition?

[flok]

Ehr right, yes, right What I meant is draw, which is somewhat the same as loosing for me

[kbhearn]

score = 0 is correct and should score best of course against similar opposition... score = -contempt would be the draw avoidance option and should provide some edge against opponents weaker than you while hurting you against similar or stronger opponents. setting contempt to mate is a wee bit extreme!

(note that -contempt needs to be from the engine's perspective not the side to move's)

[Sven]

Ehr right, yes, right What I meant is draw, which is somewhat the same as loosing for me

Are you sure that the repetition detection works correctly?

Another option might be that in the given game (against Skipper?) playing for the repetition was actually the best choice. I don't know the game so I can't judge about it.

[flok]

Ehr right, yes, right What I meant is draw, which is somewhat the same as loosing for me

Are you sure that the repetition detection works correctly?

I'm fairly certain. How I've implemented it: after doMove() I generate the TT-hash for the new position. Then I check if that hash is already in an std::unordered_set. If so: then things are repeating. If not: then store the hash and proceed.

Another option might be that in the given game (against Skipper?) playing for the repetition was actually the best choice. I don't know the game so I can't judge about it.

That was against skipper yes. I don't have the png unfortunately either but I thought maybe repetition-handling is the reason for the bad play.

[MahmoudUthman]

Do you remember to remove the hash from the set in unmake ?

[brtzsnr]

Just a reminder:

- Draw against a weaker opponent is 0.5 points lost.
- Draw against a stronger opponent is 0.5 points won.

Some engines use contempt to avoid or force a draw: https://chessprogramming.wikispaces.com/Contempt+Factor

[flok]

@mahmoud: yes I remove the hash.

@alexandru: yes but with a weak program as mine every (half-)point counts

[Henk]

Ehr right, yes, right What I meant is draw, which is somewhat the same as loosing for me

Are you sure that the repetition detection works correctly?

I'm fairly certain. How I've implemented it: after doMove() I generate the TT-hash for the new position. Then I check if that hash is already in an std::unordered_set. If so: then things are repeating. If not: then store the hash and proceed.

Another option might be that in the given game (against Skipper?) playing for the repetition was actually the best choice. I don't know the game so I can't judge about it.

That was against skipper yes. I don't have the png unfortunately either but I thought maybe repetition-handling is the reason for the bad play.

You probably mean this one. See end position. I don't think that playing for the repetition was here the best choice.

[pgn]
[Event "ICS rated blitz match"]
[Site "winboard.nl"]
[Date "2016.01.16"]
[Round "-"]
[White "Skipper"]
[Black "Embla"]
[Result "1/2-1/2"]
[WhiteElo "1120"]
[BlackElo "1051"]
[TimeControl "300+1"]
[Annotator "1. +0.00"]

1. Na3 {+0.00/26 6} e5 2. Nc4 {+0.00/19 6} e4 3. Rb1 {+0.00/18 6} d5 4. Na3
{+0.00/18 6} Nc6 5. f4 {+0.00/15 6} d4 6. g3 {+0.00/17 6} Qd5 7. Ra1
{+0.00/18 5} Bxa3 8. bxa3 {+0.20/19 5} Nf6 9. e3 {+0.20/17 5} dxe3 10. dxe3
{+0.20/16 5} Qc5 11. Ne2 {+0.20/14 5} Be6 12. Qd2 {+0.20/16 5} Rd8 13. Qc3
{+0.20/17 5} Rd5 14. Qxc5 {+0.20/18 5} Rxc5 15. c3 {+0.20/18 5} O-O 16. Bd2
{+0.20/17 5} Ra5 17. Bc1 {+0.20/18 5} Rc5 18. Bd2 {+0.00/18 5} Ra5 19. Bc1
{+0.00/20 4} Rd8 20. Rb1 {+0.20/16 4} b6 21. Ra1 {-0.40/16 4} Rc5 22. h3
{+0.20/16 5} h6 23. Bd2 {+0.20/17 5} Ra5 24. Bc1 {+0.20/17 5} Rc5 25. Bg2
{+0.20/18 5} g6 26. Kf2 {+0.20/17 5} g5 27. fxg5 {+0.20/15 5} Rf5+ 28. Nf4
{+0.20/17 5} hxg5 29. g4 {+0.20/17 5} Rc5 30. Ne2 {+0.20/16 4} Ne5 31. Kg3
{+0.00/13 4} Nd3 32. Rd1 {-0.50/14 4} Kg7 33. Nd4 {-0.60/14 4} Bd7 34. Bd2
{-0.60/14 4} Ra5 35. Rdb1 {-0.60/15 4} Rxa3 36. Nb5 {-0.60/15 4} Bxb5 37.
Rxb5 {-0.60/12 4} Nc5 38. Rb2 {-0.60/16 4} Nd3 39. Rb5 {+0.00/20 4} Nc5 40.
Rb2 {+0.00/21 4} Ra5 41. Rab1 {-0.40/16 4} Nd3 42. Rc2 {-0.40/15 2.3} Ne5
43. Be1 {-0.40/14 2.3} Nc4 44. Re2 {-0.40/14 2.3} Rd3 45. Kf2
{-0.40/14 2.2} b5 46. Ra1 {-0.40/15 2.3} Ra6 47. Rb1 {-0.40/15 2.3} Rb6 48.
Bh1 {-0.40/14 2.2} a6 49. Rc1 {-1.20/14 2.2} Re6 50. Kg1 {-1.20/15 2.2} Nd5
51. Rh2 {-1.40/13 2.1} Ndxe3 52. Bf2 {-1.20/14 2.2} c5 53. Be1
{-1.20/16 2.1} Nd5 54. h4 {-1.40/14 2.0} gxh4 55. Rxh4 {-1.40/13 2.2} Nf4
56. Kh2 {-2.00/13 2.0} Ne3 57. Rb1 {-2.00/13 2.0} f6 58. c4 {-2.00/12 2.0}
Nxc4 59. Rc1 {-2.00/13 2.1} Re8 60. Rc2 {-2.20/13 2.0} Ng6 61. Rh5
{-2.20/14 1.9} Rd1 62. Bg3 {-2.20/13 1.9} Ne3 63. Rcxc5 {-2.80/13 1.9}
Nxg4+ 64. Kh3 {-2.80/13 1.9} Rxh1+ 65. Kxg4 {-2.20/15 1.9} Rxh5 66. Rxh5
{-2.20/16 1.8} Rd8 67. Rh1 {-2.20/16 1.8} Rd2 68. Rh2 {-2.20/17 1.8} Rxh2
69. Bxh2 {-2.20/19 1.9} Ne7 70. a3 {-2.20/22 1.7} Kf7 71. Kh3
{-2.20/23 1.8} Nd5 72. Kg4 {-2.20/21 1.8} Ke6 73. Bg1 {-2.20/21 1.8} f5+
74. Kh3 {-2.20/22 1.8} e3 75. Kg2 {-2.20/17 1.7} f4 76. Bh2 {-2.20/18 1.7}
Ke5 77. Bg1 {-2.20/17 1.7} Ke4 78. Bf2 {-5.20/20 1.7} exf2 79. Kxf2
{-5.20/23 1.8} f3 80. Kg1 {-5.20/17 1.6} Ke3 81. Kh2 {-10.80/15 1.6} Nf4
82. a4 {-11.60/12 1.5} f2 83. Kg3 {-104856.40/12 1.6} bxa4 84. Kg4
{-11.60/10 1.7} f1=Q 85. Kf5 {-11.60/9 1.5} a3 86. Kf6 {-104856.40/9 1.5}
a2 87. Ke7 {-104856.80/12 1.5} a1=Q 88. Kd8 {-104857.00/14 1.6} Ne2 89. Kc7
{-104856.80/10 1.5} Qf5 90. Kb7 {-104857.00/14 1.5} Qaf6 91. Kc7
{-104857.20/19 1.5} Qd5 92. Kb8 {-104857.20/18 1.5} Ng3 93. Ka7
{-104857.20/18 1.5} Ne2 94. Kb8 {+0.00/200 0.7} Ng3 95. Ka7 {+0.00/200 1.1}
Ne2 96. Kb8 {+0.00/200 0.2}
{Game drawn by repetition} 1/2-1/2
[/pgn]