960 rematch using Komodo MCTS

[lkaufman]

In honor of the 960 matches now in progress at the St. Louis chess club, this Wednesday (Sep 4) evening I plan to repeat my960 rapid match vs. Komodo at knight odds, with the only difference being that I will play against Komodo MCTS, instead of against normal Komodo. The opening positions, time limit, hardware, etc. will all be the same. We really have no data on whether normal Komodo or Komodo MCTS is tougher for human opponents, this will be the first direct comparison. To avoid conflicting with the St. Louis matches, I'll start at 7:30 p.m. Easter time on this occasion. While normally I would expect the MCTS version to be more "tricky" for a human opponent, even if it is slightly weaker objectively, this may not be the case for these knight odds games, because currently Contempt only works with normal Komodo. Komodo MCTS effectively has some built-in Contempt due to the nature of the MCTS algorithm, but it may not be enough for knight odds. Anyway, we'll see. Of course four games is too small a sample for the results to be statistically significant, but perhaps the actual games will tell us something beyond what the results reveal.

[lkaufman]

Starting in about an hour and twenty minutes on chess.com, live.

[lkaufman]

I got in big trouble in just the first few moves.

[pgn][Event "Live Chess - 960 knight odds"] [Site "Chess.com"] [Date "2019.09.04"] [Round "1"] [White "PlayKomodo (Komodo MCTS)"] [Black "hissha (GM Larry Kaufman)"] [Result "1-0"] [SetUp "1"] [FEN "rnqbknbr/pppppppp/8/8/8/8/PPPPPPPP/R1QBKNBR w KQkq - 0 1"] [TimeControl "900+10"] [WhiteElo "2960"] [BlackElo "2294"] [Termination "PlayKomodo won by resignation"] 1. f4 {[%clk 0:15:07.4]} 1... f5 {[%clk 0:14:52.3]} 2. Ng3 {[%clk 0:15:12.4]} 2... d6 {[%clk 0:14:30.1]} 3. e4 {[%clk 0:15:22.3]} 3... fxe4 {[%clk 0:14:23]} 4. Nh5 {[%clk 0:15:08.2]} 4... Bd5 {[%clk 0:10:20.3]} 5. c4 {[%clk 0:14:23.2]} 5... Bc6 {[%clk 0:10:04.4]} 6. Nxg7+ {[%clk 0:13:30.7]} 6... Kf7 {[%clk 0:10:07.5]} 7. Bd4 {[%clk 0:12:53.6]} 7... e5 {[%clk 0:07:50.2]} 8. fxe5 {[%clk 0:12:36]} 8... Kxg7 {[%clk 0:07:57.3]} 9. exd6+ {[%clk 0:11:36.6]} 9... Kg8 {[%clk 0:07:54.9]} 10. O-O {[%clk 0:09:31.4]} 10... Ne6 {[%clk 0:05:31.8]} 11. Qc3 {[%clk 0:09:17.8]} 11... cxd6 {[%clk 0:04:01.6]} 12. Bh5 {[%clk 0:09:27.7]} 12... Nxd4 {[%clk 0:02:25.7]} 13. Qg3+ [/pgn]

[lkaufman]

This was a much better game. I had to play well to force a draw.

[pgn][Event "Live Chess - 960 knight odds"] [Site "Chess.com"] [Date "2019.09.04"] [Round "2"] [White "PlayKomodo (Komodo mcts)"] [Black "hissha (GM Larry Kaufman"] [Result "1/2-1/2"] [SetUp "1"] [FEN "rnbnkbqr/pppppppp/8/8/8/8/PPPPPPPP/R1BNKBQR w KQkq - 0 1"] [TimeControl "900+10"] [WhiteElo "2960"] [BlackElo "2294"] [Termination "Game drawn by repetition"] 1. f4 {[%clk 0:15:07.4]} 1... d5 {[%clk 0:13:55.9]} 2. e4 {[%clk 0:15:12.4]} 2... dxe4 {[%clk 0:13:30]} 3. d3 {[%clk 0:15:22.3]} 3... exd3 {[%clk 0:12:41.2]} 4. Bxd3 {[%clk 0:15:27.6]} 4... g6 {[%clk 0:11:33.1]} 5. Nc3 {[%clk 0:15:37.5]} 5... c6 {[%clk 0:10:34.5]} 6. Be3 {[%clk 0:14:43.3]} 6... Bg7 {[%clk 0:09:59.5]} 7. O-O-O {[%clk 0:14:27.7]} 7... Bg4 {[%clk 0:08:54.2]} 8. Ne2 {[%clk 0:13:40.3]} 8... f5 {[%clk 0:07:46.8]} 9. Kb1 {[%clk 0:13:13]} 9... Qf7 {[%clk 0:07:24.5]} 10. Rd2 {[%clk 0:11:11.5]} 10... Bxe2 {[%clk 0:07:19.6]} 11. Rxe2 {[%clk 0:10:12.4]} 11... O-O {[%clk 0:05:44.3]} 12. g4 {[%clk 0:09:53.5]} 12... fxg4 {[%clk 0:05:25.8]} 13. Qxg4 {[%clk 0:08:15.6]} 13... Na6 {[%clk 0:05:13.7]} 14. c3 {[%clk 0:08:01]} 14... Nc7 {[%clk 0:04:37.4]} 15. Bc2 {[%clk 0:08:03.1]} 15... Nd5 {[%clk 0:04:20.6]} 16. Bc1 {[%clk 0:07:47.1]} 16... e6 {[%clk 0:03:54.2]} 17. h4 {[%clk 0:07:56.9]} 17... Qf6 {[%clk 0:02:55.9]} 18. h5 {[%clk 0:07:37.7]} 18... Nxc3+ {[%clk 0:01:47.5]} 19. bxc3 {[%clk 0:07:47.6]} 19... Qxc3 {[%clk 0:01:53.6]} 20. Bb3 {[%clk 0:07:27.9]} 20... Qa1+ {[%clk 0:01:36.3]} 21. Kc2 {[%clk 0:07:37.8]} 21... Qc3+ {[%clk 0:01:38.5]} 22. Kb1 {[%clk 0:06:57.3]} 22... Qa1+ {[%clk 0:01:34.4]} 23. Kc2 {[%clk 0:07:07.2]} 23... Qc3+ {[%clk 0:01:38.7]} 24. Kb1 {[%clk 0:03:59.4]} 1/2-1/2[/pgn]

[lkaufman]

I managed to exchange queens while still a knight ahead, but the passed pawn was difficult to deal with, and with little time I couldn't find the right way.

[pgn][Event "Live Chess - 960 knight odds"] [Site "Chess.com"] [Date "2019.09.04"] [Round "3"] [White "PlayKomodo (komodo mcts)"] [Black "hissha (GM Larry Kaufman)"] [Result "1-0"] [SetUp "1"] [FEN "rnbbknqr/pppppppp/8/8/8/8/PPPPPPPP/R1BBKNQR w KQkq - 0 1"] [TimeControl "900+10"] [WhiteElo "2960"] [BlackElo "2294"] [Termination "PlayKomodo won on time"] 1. d4 {[%clk 0:15:07.4]} 1... d5 {[%clk 0:14:51.9]} 2. c3 {[%clk 0:15:17.3]} 2... f5 {[%clk 0:13:57.3]} 3. Bc2 {[%clk 0:14:21.1]} 3... Qf7 {[%clk 0:13:48.7]} 4. h4 {[%clk 0:13:45.1]} 4... Nbd7 {[%clk 0:13:21.7]} 5. f3 {[%clk 0:13:26.2]} 5... Nf6 {[%clk 0:13:10]} 6. Ng3 {[%clk 0:11:54.4]} 6... e6 {[%clk 0:13:02.4]} 7. e4 {[%clk 0:10:24.2]} 7... fxe4 {[%clk 0:12:28.4]} 8. fxe4 {[%clk 0:09:50.5]} 8... Nxe4 {[%clk 0:12:26.9]} 9. Nxe4 {[%clk 0:09:28.4]} 9... dxe4 {[%clk 0:12:36.8]} 10. Qe3 {[%clk 0:08:12]} 10... Nd7 {[%clk 0:12:15.1]} 11. Qxe4 {[%clk 0:08:21.9]} 11... Nf6 {[%clk 0:11:53.6]} 12. Qd3 {[%clk 0:07:49.6]} 12... Bd7 {[%clk 0:10:26.8]} 13. O-O {[%clk 0:07:59.5]} 13... Qh5 {[%clk 0:08:57.7]} 14. Bg5 {[%clk 0:07:31.8]} 14... Be7 {[%clk 0:08:20.3]} 15. Rae1 {[%clk 0:07:33.7]} 15... O-O-O {[%clk 0:07:56.5]} 16. c4 {[%clk 0:07:22.7]} 16... h6 {[%clk 0:07:26.8]} 17. Bd1 {[%clk 0:07:00.1]} 17... Qf7 {[%clk 0:07:09]} 18. Bd2 {[%clk 0:05:59.3]} 18... g5 {[%clk 0:06:38.9]} 19. h5 {[%clk 0:04:47.9]} 19... Rhf8 {[%clk 0:06:35.9]} 20. Qg6 {[%clk 0:03:55.3]} 20... Qxg6 {[%clk 0:05:52.6]} 21. hxg6 {[%clk 0:03:49.1]} 21... Ng8 {[%clk 0:03:09.1]} 22. Rf7 {[%clk 0:03:24.8]} 22... Bf6 {[%clk 0:01:21]} 23. Bc3 {[%clk 0:03:19]} 23... c5 {[%clk 0:00:54]} 24. Rf1 {[%clk 0:03:06.4]} 24... cxd4 {[%clk 0:00:36.2]} 25. Bb4 {[%clk 0:02:47.5]} 25... Rfe8 {[%clk 0:00:17.3]} 26. R1xf6 {[%clk 0:02:28.5]} 26... Nxf6 {[%clk 0:00:27.2]} 27. Rxf6 {[%clk 0:02:20.7]} 27... Rg8 {[%clk 0:00:27.7]} 28. Bh5 {[%clk 0:02:15.5]} 28... e5 {[%clk 0:00:30.9]} 29. Kf2 {[%clk 0:01:50.5]} 29... e4 {[%clk 0:00:31.1]} 30. Bc5 {[%clk 0:01:21.2]} 30... e3+ {[%clk 0:00:31.7]} 31. Ke1 {[%clk 0:01:09.2]} 31... Bc6 {[%clk 0:00:33]} 32. Bxa7 {[%clk 0:00:51.9]} 32... d3 {[%clk 0:00:25]} 33. Bxe3 {[%clk 0:00:42.1]} 33... Rge8 {[%clk 0:00:28.5]} 34. Kd2 {[%clk 0:00:35.7]} 34... Bxg2 {[%clk 0:00:23.7]} 35. Bg4+ {[%clk 0:00:29.8]} 35... Kb8 {[%clk 0:00:25.1]} 36. Bb6 {[%clk 0:00:34.3]} 36... Rc8 {[%clk 0:00:20.5]} 37. Bxc8 {[%clk 0:00:18.8]} 37... Kxc8 {[%clk 0:00:26.7]} 38. g7 {[%clk 0:00:21.1]} 38... Kd7 {[%clk 0:00:21.2]} 39. Rf8 {[%clk 0:00:31]} 39... Re2+ {[%clk 0:00:21.7]} 40. Kc3 {[%clk 0:00:21.5]} 1-0[/pgn]

[lkaufman]

This time a fairly routine win, Komodo getting just slight compensation for the knight. So I made the same score as against normal Komodo, 1.5 out of 4. My impression is that MCTS needs Contempt to do well at knight odds; its built-in contempt is not nearly enough for this handicap.

[pgn][Event "Live Chess - 960 knight odds"] [Site "Chess.com"] [Date "2019.09.04"] [Round "4"] [White "PlayKomodo(Komodo MCTS)"] [Black "hissha (GM Larry Kaufman)"] [Result "0-1"] [SetUp "1"] [FEN "rnqnkbbr/pppppppp/8/8/8/8/PPPPPPPP/R1QNKBBR w KQkq - 0 1"] [TimeControl "900+10"] [WhiteElo "2960"] [BlackElo "2294"] [Termination "hissha won by checkmate"] 1. d4 {[%clk 0:15:07.4]} 1... f5 {[%clk 0:14:35.3]} 2. f3 {[%clk 0:15:17.3]} 2... g6 {[%clk 0:14:16.9]} 3. e4 {[%clk 0:14:21.4]} 3... fxe4 {[%clk 0:14:19.9]} 4. fxe4 {[%clk 0:13:57.8]} 4... Bg7 {[%clk 0:14:26.6]} 5. Qd2 {[%clk 0:12:40.4]} 5... Bf7 {[%clk 0:14:16.2]} 6. h4 {[%clk 0:11:12.2]} 6... d6 {[%clk 0:13:52.7]} 7. h5 {[%clk 0:10:55.9]} 7... gxh5 {[%clk 0:12:38.6]} 8. Ne3 {[%clk 0:11:05.8]} 8... e6 {[%clk 0:12:05.3]} 9. Be2 {[%clk 0:10:13]} 9... Nd7 {[%clk 0:11:52.8]} 10. O-O-O {[%clk 0:09:12.8]} 10... Nf6 {[%clk 0:11:32.9]} 11. e5 {[%clk 0:08:56.8]} 11... Ne4 {[%clk 0:10:41.2]} 12. Qe1 {[%clk 0:09:06.7]} 12... d5 {[%clk 0:10:28.4]} 13. Bxh5 {[%clk 0:07:46.6]} 13... Bxh5 {[%clk 0:10:04.4]} 14. Rxh5 {[%clk 0:07:17.8]} 14... Nf7 {[%clk 0:10:11.1]} 15. c4 {[%clk 0:05:13.3]} 15... c6 {[%clk 0:09:09.6]} 16. Kb1 {[%clk 0:05:23.2]} 16... Qd8 {[%clk 0:08:21.7]} 17. Ng4 {[%clk 0:04:53.6]} 17... Qb6 {[%clk 0:06:27.9]} 18. Be3 {[%clk 0:04:29.5]} 18... O-O-O {[%clk 0:06:00.6]} 19. Nf2 {[%clk 0:04:00.3]} 19... Nxf2 {[%clk 0:05:43.5]} 20. Qxf2 {[%clk 0:03:30.3]} 20... Rdf8 {[%clk 0:05:41]} 21. Qe2 {[%clk 0:02:52.5]} 21... Nd8 {[%clk 0:05:13.4]} 22. cxd5 {[%clk 0:02:54.2]} 22... cxd5 {[%clk 0:04:58.8]} 23. Rdh1 {[%clk 0:02:55.8]} 23... h6 {[%clk 0:04:43.7]} 24. Rc1+ {[%clk 0:02:58.7]} 24... Kb8 {[%clk 0:04:39.2]} 25. Rh3 {[%clk 0:02:51.6]} 25... Rf7 {[%clk 0:04:12.6]} 26. Rf1 {[%clk 0:02:29.3]} 26... Rxf1+ {[%clk 0:04:08.7]} 27. Qxf1 {[%clk 0:02:18.6]} 27... Rf8 {[%clk 0:04:14.5]} 28. Qd3 {[%clk 0:01:13.8]} 28... Qa6 {[%clk 0:03:22]} 29. Qxa6 {[%clk 0:01:12.3]} 29... bxa6 {[%clk 0:03:31.9]} 30. Kc2 {[%clk 0:00:52.3]} 30... Kc8 {[%clk 0:02:59.8]} 31. Rg3 {[%clk 0:00:46.4]} 31... Rf7 {[%clk 0:02:53.5]} 32. Kd3 {[%clk 0:00:54.6]} 32... Rb7 {[%clk 0:02:33.3]} 33. Rg6 {[%clk 0:00:46.9]} 33... a5 {[%clk 0:02:22.6]} 34. Bd2 {[%clk 0:00:43.5]} 34... a4 {[%clk 0:02:18.9]} 35. Ba5 {[%clk 0:00:47.8]} 35... Bf8 {[%clk 0:02:10.3]} 36. Rf6 {[%clk 0:00:57.7]} 36... Be7 {[%clk 0:02:09.7]} 37. Rf2 {[%clk 0:01:01.3]} 37... Nc6 {[%clk 0:02:07.7]} 38. Bc3 {[%clk 0:00:52.3]} 38... Kd7 {[%clk 0:02:03.3]} 39. Kc2 {[%clk 0:00:50.4]} 39... Rb8 {[%clk 0:01:57.6]} 40. Rf7 {[%clk 0:00:58.9]} 40... Rf8 {[%clk 0:01:55.2]} 41. Rg7 {[%clk 0:00:49.3]} 41... Rf2+ {[%clk 0:01:57.7]} 42. Kd3 {[%clk 0:00:33.7]} 42... a3 {[%clk 0:01:53]} 43. bxa3 {[%clk 0:00:31.8]} 43... Rxa2 {[%clk 0:01:53.9]} 44. Bb4 {[%clk 0:00:30.4]} 44... Nxb4+ {[%clk 0:01:51.7]} 45. axb4 {[%clk 0:00:40.3]} 45... a6 {[%clk 0:01:41]} 46. b5 {[%clk 0:00:34.9]} 46... axb5 {[%clk 0:01:46.8]} 47. Rg8 {[%clk 0:00:34.8]} 47... h5 {[%clk 0:01:46.7]} 48. g4 {[%clk 0:00:27.2]} 48... Ra3+ {[%clk 0:01:45]} 49. Kc2 {[%clk 0:00:21.1]} 49... h4 {[%clk 0:01:41.8]} 50. g5 {[%clk 0:00:19.9]} 50... Rg3 {[%clk 0:01:34.5]} 51. g6 {[%clk 0:00:19.4]} 51... b4 {[%clk 0:01:30.9]} 52. g7 {[%clk 0:00:29.3]} 52... h3 {[%clk 0:00:55.5]} 53. Kb2 {[%clk 0:00:25.7]} 53... h2 {[%clk 0:00:56.2]} 54. Rh8 {[%clk 0:00:27]} 54... Rxg7 {[%clk 0:00:59.7]} 55. Rxh2 {[%clk 0:00:29.7]} 55... Rg4 {[%clk 0:01:04.4]} 56. Rh7 {[%clk 0:00:13.4]} 56... Rxd4 {[%clk 0:01:07.6]} 57. Rg7 {[%clk 0:00:20.1]} 57... Re4 {[%clk 0:01:10.4]} 58. Rg2 {[%clk 0:00:13.6]} 58... Rxe5 {[%clk 0:01:16.8]} 59. Kc2 {[%clk 0:00:12]} 59... Re3 {[%clk 0:01:21.6]} 60. Rg7 {[%clk 0:00:12.3]} 60... b3+ {[%clk 0:01:25.8]} 61. Kd2 {[%clk 0:00:12.5]} 61... Rh3 {[%clk 0:01:20.6]} 62. Rg4 {[%clk 0:00:18]} 62... Rh2+ {[%clk 0:01:18.5]} 63. Kc3 {[%clk 0:00:16.9]} 63... b2 {[%clk 0:01:18.3]} 64. Rg1 {[%clk 0:00:22.4]} 64... e5 {[%clk 0:01:21.5]} 65. Rh1 {[%clk 0:00:14.9]} 65... Rxh1 {[%clk 0:01:23.8]} 66. Kb3 {[%clk 0:00:22.2]} 66... b1=Q+ {[%clk 0:01:29.2]} 67. Kc3 {[%clk 0:00:14.1]} 67... Bb4# {[%clk 0:01:25.4]} 0-1[/pgn]

[JJJ]

I like you doing these match Larry. Thanks you for the show.

[lkaufman]

I like you doing these match Larry. Thanks you for the show.

We learn things from them. In this case my impression is that although MCTS is probably more difficult for a human to play against than normal Komodo, with large handicaps we really do need Contempt to work with MCTS, and so have restored it for future matches. I need to test what values make sense for MCTS.
I want to add that having watched the St. Louis elite matches of chess 960 this week as well as having played two 960 matches myself with Komodo, I am liking the 960 game more and more. I think it may have a big role in the future of chess. Chess is supposed to be a thinking game, not a memory contest. With the World 960 championship coming soon, we'll see how interest grows. It is especially suitable for engine events, since there is no need to decide on preselected openings. For handicap play, we are pretty much limited to rook for knight, knight odds, and rook odds, since the value of pawns and/or tempi are highly position-dependent. But those three (actually six since each can be with the engine White (standard) or Black (+move)) should be enough.