Mythbuster continued

Discussion of computer chess matches and engine tournaments.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
beram
Posts: 1187
Joined: Wed Jan 06, 2010 2:11 pm

Mythbuster continued

Post by beram » Thu Mar 08, 2012 8:03 pm

In the great Clash of the Titans matches of Timo Klaustermeyer (see below and see http://www.team-oh.de/Computerschach/Clash.htm )
with long time control 125min / 50 moves + (60min+5s) -
Houdini2.0C leads against Komodo4 with 6-4.
It won two games as white and black after same French opening. So far the match looks like another mythbuster.
Interesting is the following game where Houdini wins easily with black.

after following position
[D] 2kr4/p3np2/Pqb1p2r/1p2Pn1B/3pNP2/2pQ4/2P2RPP/R1B3K1 w - - 0 24

Komodo plays - after 6min 48 sec, at depth 22 - the bad move 24. Nf6 where after Rhd8 gave black a winning position. Giving Komodo4 more time it would perhaps find the better Bf3 after which he still is in the game. Komodo 3 which is (almost) equally strong as Komodo 4, prefers on my PC 24. Bf3 already at search depth 20 and stays with it till depth 23 giving black -0.32 with a line beginning with 24. ..Nd5 25. Rb1 - b4
Hereby the whole game:

(8) Komodo64 SSE Version 4 - Houdini 2.0c Pro x64 [t=1] [C18]
Houdini-Komodo (8.1), 08.03.2012
[0.22;0.25]
AMD Phenom(tm) II X4 905e Processor 2999 MHz W=22.5 plies; 908kN/s B=22.0 plies; 2.062kN/s; 6.159 TBAs 1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.e5 c5 5.a3 Bxc3+ 6.bxc3 Qc7 7.Qg4 Ne7 8.Qxg7 Rg8 9.Qxh7 cxd4 10.Ne2 Nbc6 11.f4 dxc3 12.Qd3 d4 13.Ng3 0.22/23 5:03 Bd7 0.25/23 3:58 14.Be2 (Ne4) 0.24/24 3:53 14...Qb6 -0.09/23 2:56 15.0–0 0.38/25 22 0–0–0 -0.09/24 2:19 16.Ne4 0.23/24 1:44 Nf5 (Kb8) -0.10/23 1:35 17.Nf6 0.50/23 3:34 Rg6 -0.12/22 0 18.Rf2 (g3) 0.33/22 3:04 18...Nce7 (a5) -0.26/23 8:03 19.a4 (Nxd7) 0.20/22 3:08 19...Bc6 (Qa5) -0.31/23 6:40 20.a5 -0.05/22 3:48 Qc5 (Qc7) -0.41/22 2:14 21.a6 0.04/22 3:46 b5 -0.37/24 0 22.Ne4 -0.05/22 3:11 Qb6 -0.45/23 0 23.Bh5 -0.05/22 5:44 Rh6 (Rg7) -0.72/24 17:30 24.Nf6 (Bf3) -0.26/22 6:48 [Komodo64 3: 24.Bf3 Nd5 25.Rb1 b4 26.Qc4 d3 27.cxd3 Nd4 28.Kh1 Kb8 29.f5 c2 30.Rb2 Rh4 31.g3 Rhh8 32.fxe6 fxe6 33.Rbxc2 Bb5 34.Qc5 Nxc2 35.Qxc2 Rc8 36.Qd2 Qxa6 37.Nd6 Rc3 38.Bxd5 exd5 -0.32/23 ] 24...Rdh8 (Rxh5) -1.26/20 2:47 25.g4 -0.67/22 2:43 Nd5 (Ba8) -2.90/21 8:11 26.Bxf7 -1.54/21 4:38 Nb4 (Nxf6) -3.44/22 0 27.Bxe6+ -2.11/21 3:36 Kd8 -4.29/22 6:38 28.Bxf5 -1.96/23 2:46 Nxd3 -4.19/22 3:29 29.Bxd3 -2.53/23 2:26 Qc5 (Rh3) -3.58/23 13:34 30.Re2 -1.57/23 2:37 Rh3 (b4) -3.25/22 8:54 31.e6 -2.31/21 5:50 Ke7 -3.67/24 3:28 32.Ne4 (Re5) -2.80/23 8:43 32...Bxe4 -5.87/17 49 33.Rxe4 -3.05/22 42 Rg8 -5.70/19 0 34.f5 -3.23/22 2:06 Rxd3 -5.83/21 0 35.f6+ -4.27/22 6:03 Kxf6 -6.16/23 0 36.cxd3 -3.93/20 14 Qh5 -5.96/22 0 37.g5+ -4.24/24 3:18 Ke7 -6.11/24 0 38.Ba3+ -3.80/25 4:10 Ke8 -6.73/25 0 39.h4 -4.28/26 3:35 Qf3 -6.77/25 0 40.Rf1 -4.55/25 3:59 Qxd3 -7.06/24 0 41.Ref4 -4.78/25 3:28 Qg3+ -7.06/24 0 42.Kh1 -4.77/22 8 Qh3+ -7.13/22 1:04 43.Kg1 -4.78/25 2:30 Qxe6 -7.42/23 3:18 44.R1f3 -4.77/25 1:41 Qe2 -7.42/23 0 45.Rf2 -5.24/24 3:10 Qe1+ (Qe5) -7.73/22 0 46.Kg2 -4.43/23 48 d3 -7.80/21 42 47.Bb4 -4.39/24 15 d2 (Qe5) -7.94/20 40 48.Bxc3 -4.16/21 38 Qxf2+ -7.90/21 0 49.Kxf2 -4.16/20 21 d1Q -7.02/21 0 50.Bd4 -4.72/23 5:02 Qc1 (Qc2+) -8.21/22 0 51.Kf3 (Kg3) -5.01/22 2:53 51...b4 (Qa3+) -9.41/17 51 52.Bxa7 (Be3) -4.47/18 1:28 52...b3 (Qf1+) -10.82/17 44 53.Bd4 -5.10/19 1:56 b2 (Qf1+) -11.82/21 0 54.Bxb2 -6.18/17 37 Qxb2 -17.33/20 2:10 0–1

grts Bram

pichy
Posts: 2564
Joined: Thu Mar 09, 2006 2:04 am

Re: I have a recommendation for you and all testers

Post by pichy » Fri Mar 09, 2012 12:42 am

beram wrote:In the great Clash of the Titans matches of Timo Klaustermeyer (see below and see http://www.team-oh.de/Computerschach/Clash.htm )
with long time control 125min / 50 moves + (60min+5s) -
Houdini2.0C leads against Komodo4 with 6-4.
It won two games as white and black after same French opening. So far the match looks like another mythbuster.
Interesting is the following game where Houdini wins easily with black.

after following position
[D] 2kr4/p3np2/Pqb1p2r/1p2Pn1B/3pNP2/2pQ4/2P2RPP/R1B3K1 w - - 0 24

Komodo plays - after 6min 48 sec, at depth 22 - the bad move 24. Nf6 where after Rhd8 gave black a winning position. Giving Komodo4 more time it would perhaps find the better Bf3 after which he still is in the game. Komodo 3 which is (almost) equally strong as Komodo 4, prefers on my PC 24. Bf3 already at search depth 20 and stays with it till depth 23 giving black -0.32 with a line beginning with 24. ..Nd5 25. Rb1 - b4
Hereby the whole game:

(8) Komodo64 SSE Version 4 - Houdini 2.0c Pro x64 [t=1] [C18]
Houdini-Komodo (8.1), 08.03.2012
[0.22;0.25]
AMD Phenom(tm) II X4 905e Processor 2999 MHz W=22.5 plies; 908kN/s B=22.0 plies; 2.062kN/s; 6.159 TBAs 1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.e5 c5 5.a3 Bxc3+ 6.bxc3 Qc7 7.Qg4 Ne7 8.Qxg7 Rg8 9.Qxh7 cxd4 10.Ne2 Nbc6 11.f4 dxc3 12.Qd3 d4 13.Ng3 0.22/23 5:03 Bd7 0.25/23 3:58 14.Be2 (Ne4) 0.24/24 3:53 14...Qb6 -0.09/23 2:56 15.0–0 0.38/25 22 0–0–0 -0.09/24 2:19 16.Ne4 0.23/24 1:44 Nf5 (Kb8) -0.10/23 1:35 17.Nf6 0.50/23 3:34 Rg6 -0.12/22 0 18.Rf2 (g3) 0.33/22 3:04 18...Nce7 (a5) -0.26/23 8:03 19.a4 (Nxd7) 0.20/22 3:08 19...Bc6 (Qa5) -0.31/23 6:40 20.a5 -0.05/22 3:48 Qc5 (Qc7) -0.41/22 2:14 21.a6 0.04/22 3:46 b5 -0.37/24 0 22.Ne4 -0.05/22 3:11 Qb6 -0.45/23 0 23.Bh5 -0.05/22 5:44 Rh6 (Rg7) -0.72/24 17:30 24.Nf6 (Bf3) -0.26/22 6:48 [Komodo64 3: 24.Bf3 Nd5 25.Rb1 b4 26.Qc4 d3 27.cxd3 Nd4 28.Kh1 Kb8 29.f5 c2 30.Rb2 Rh4 31.g3 Rhh8 32.fxe6 fxe6 33.Rbxc2 Bb5 34.Qc5 Nxc2 35.Qxc2 Rc8 36.Qd2 Qxa6 37.Nd6 Rc3 38.Bxd5 exd5 -0.32/23 ] 24...Rdh8 (Rxh5) -1.26/20 2:47 25.g4 -0.67/22 2:43 Nd5 (Ba8) -2.90/21 8:11 26.Bxf7 -1.54/21 4:38 Nb4 (Nxf6) -3.44/22 0 27.Bxe6+ -2.11/21 3:36 Kd8 -4.29/22 6:38 28.Bxf5 -1.96/23 2:46 Nxd3 -4.19/22 3:29 29.Bxd3 -2.53/23 2:26 Qc5 (Rh3) -3.58/23 13:34 30.Re2 -1.57/23 2:37 Rh3 (b4) -3.25/22 8:54 31.e6 -2.31/21 5:50 Ke7 -3.67/24 3:28 32.Ne4 (Re5) -2.80/23 8:43 32...Bxe4 -5.87/17 49 33.Rxe4 -3.05/22 42 Rg8 -5.70/19 0 34.f5 -3.23/22 2:06 Rxd3 -5.83/21 0 35.f6+ -4.27/22 6:03 Kxf6 -6.16/23 0 36.cxd3 -3.93/20 14 Qh5 -5.96/22 0 37.g5+ -4.24/24 3:18 Ke7 -6.11/24 0 38.Ba3+ -3.80/25 4:10 Ke8 -6.73/25 0 39.h4 -4.28/26 3:35 Qf3 -6.77/25 0 40.Rf1 -4.55/25 3:59 Qxd3 -7.06/24 0 41.Ref4 -4.78/25 3:28 Qg3+ -7.06/24 0 42.Kh1 -4.77/22 8 Qh3+ -7.13/22 1:04 43.Kg1 -4.78/25 2:30 Qxe6 -7.42/23 3:18 44.R1f3 -4.77/25 1:41 Qe2 -7.42/23 0 45.Rf2 -5.24/24 3:10 Qe1+ (Qe5) -7.73/22 0 46.Kg2 -4.43/23 48 d3 -7.80/21 42 47.Bb4 -4.39/24 15 d2 (Qe5) -7.94/20 40 48.Bxc3 -4.16/21 38 Qxf2+ -7.90/21 0 49.Kxf2 -4.16/20 21 d1Q -7.02/21 0 50.Bd4 -4.72/23 5:02 Qc1 (Qc2+) -8.21/22 0 51.Kf3 (Kg3) -5.01/22 2:53 51...b4 (Qa3+) -9.41/17 51 52.Bxa7 (Be3) -4.47/18 1:28 52...b3 (Qf1+) -10.82/17 44 53.Bd4 -5.10/19 1:56 b2 (Qf1+) -11.82/21 0 54.Bxb2 -6.18/17 37 Qxb2 -17.33/20 2:10 0–1

grts Bram

Instead of using a time control where you have to make certain moves within X amount of time, with long time control 125min / 50 moves + (60min+5s) - simply give each engine 8 minutes per move. For human it is better to use the old X amount of moves in X amount of time, since we can manage the time better than engines. for instance in a well known opening that it is equal we don't spend too much time. I truly believe that providing X amount of time per move is a better method to test two engines of similar strength.

User avatar
Matthias Gemuh
Posts: 3238
Joined: Thu Mar 09, 2006 8:10 am
Contact:

Re: I have a recommendation for you and all testers

Post by Matthias Gemuh » Fri Mar 09, 2012 6:04 am

pichy wrote:
Instead of using a time control where you have to make certain moves within X amount of time, with long time control 125min / 50 moves + (60min+5s) - simply give each engine 8 minutes per move. For human it is better to use the old X amount of moves in X amount of time, since we can manage the time better than engines. for instance in a well known opening that it is equal we don't spend too much time. I truly believe that providing X amount of time per move is a better method to test two engines of similar strength.
If an engine is poor at managing time, testers should expose the problem so that it can be fixed.
Handicapping the opponent engine to achieve equal time management, is a bad thing to do.
"X amount of time per move" is capital punishment for all smart engines.

Matthias.
My engine was quite strong till I added knowledge to it.
http://www.chess.hylogic.de

IWB
Posts: 1539
Joined: Thu Mar 09, 2006 1:02 pm

Re: I have a recommendation for you and all testers

Post by IWB » Fri Mar 09, 2012 6:19 am

Matthias Gemuh wrote:
pichy wrote:
Instead of using a time control where you have to make certain moves within X amount of time, with long time control 125min / 50 moves + (60min+5s) - simply give each engine 8 minutes per move. For human it is better to use the old X amount of moves in X amount of time, since we can manage the time better than engines. for instance in a well known opening that it is equal we don't spend too much time. I truly believe that providing X amount of time per move is a better method to test two engines of similar strength.
If an engine is poor at managing time, testers should expose the problem so that it can be fixed.
Handicapping the opponent engine to achieve equal time management, is a bad thing to do.
"X amount of time per move" is capital punishment for all smart engines.

Matthias.
Hello Mathias,

I fully agree. Helping an engine with the time control is directly influencing with the capability of the programmer to write a decent time control ...
Unfortunately playing "certain moves within X amount of time" and repeating that is doing exactly the same (and it is just copying a human habit as humans have problems with time as well or even worse than engines).
The only fair and logical time control would be "Game in x time" - done! If the programm splits that x in three parts and plays 40 moves in X/3 + 40moves in X/3 + Rest in x/3 or divides it by 1.5 times the number of assumed moves or is doing something more sophisticated is completly in the hand of the programmer than.

If I would have to play the IPON again I would definitly go for a Game/x! (+ a very short increment (1s) to avoid losses on time because of a possible network lag).

Bye
Ingo

TimoK
Posts: 97
Joined: Sun Jan 03, 2010 11:28 am
Location: Hamburg

Re: I have a recommendation for you and all testers

Post by TimoK » Fri Mar 09, 2012 6:40 am

Hi Ingo,

playing "certain moves within X amount of time" is a totally different thing as "X amount of time per move". But I guess you realized the difference. Of course, the TC "game in x" is giving best independence to the engines time management. But at some tournaments (not only human, but also computerchess like former ICGA WCCCs) engines have to fulfill the constraint to make X moves in Y time. So that is (computer)chess practice, too. Of course I agree to your argument that this TCs can help engines with less smart time management. If you want to test how clever an engine's time management is, you must test "game in x".

Best regards
Timo

beram
Posts: 1187
Joined: Wed Jan 06, 2010 2:11 pm

Re: Mythbuster continued

Post by beram » Fri Mar 09, 2012 7:13 am

and it goes on:

Houdini-Komodo 2012

1 Houdini 2.0c Pro x64 [t=1] ½½½01½11½½1½1??????????????????????????????????????????????? 8.5 / 13
2 Komodo64 SSE Version 4 ½½½10½00½½0½0??????????????????????????????????????????????? 4.5 / 13

IWB
Posts: 1539
Joined: Thu Mar 09, 2006 1:02 pm

Re: I have a recommendation for you and all testers

Post by IWB » Fri Mar 09, 2012 7:30 am

TimoK wrote:
playing "certain moves within X amount of time" is a totally different thing as "X amount of time per move". But I guess you realized the difference.
Yes, of course - and both is influencing the engine ...

TimoK wrote: ... But at some tournaments (not only human, but also computerchess like former ICGA WCCCs) engines have to fulfill the constraint to make X moves in Y time. So that is (computer)chess practice, too....
You are right, but that doesn't mean that it is right ...
(But the last years usually they played game in x time)

As you just agreed, the least influence in the complete chess entity (aka "engine") is "Game in x time".

Regards
Ingo

Jouni
Posts: 2029
Joined: Wed Mar 08, 2006 7:15 pm

Re: Mythbuster continued

Post by Jouni » Fri Mar 09, 2012 7:39 am

Also Stockfish is in trouble, but Critter and Rybka have both at leat 50%! OK still early days.
Jouni

beram
Posts: 1187
Joined: Wed Jan 06, 2010 2:11 pm

Re: Mythbuster continued ...but Komodo comes back

Post by beram » Fri Mar 09, 2012 9:39 pm

Komodo has come back winning two games, one was game 13 with a favorable for white opening, a Pirc defense, Austrian attack (a white win by both)
But game 16 was a clean cut victory. After interesting opening play after move 15, [D] r2qkb1r/1b1n1ppp/p2Pp3/1p2P3/2p2B2/5N2/PP2QPPP/1K1R1B1R b kq - 0 15 Houdini with black was way too optimistic in his evals (15…Rc8 -0,06 while Komodo answered very nice with 16.h4 ! +0,55) Komodo had a passed pawn at d6, while Houdini hadn’t developed his kingside yet. Probably Houdini reckoned on compensation against the opposite white king, which had castled long missing the c-pawn. After move 21. Komodo semiforced exchange of the active black rook leaving black with a position where he was too far behind in development and Komodo still had his free d-pawn. The rest of the game Komodo demonstrated clean technique.
So now after 16 games, 9-7 for Houdini +5-3=8

(16) Komodo64 SSE Version 4 - Houdini 2.0c Pro x64 [t=1] [D43]
Houdini-Komodo (16.1), 09.03.2012
[0.27;0.11]
AMD Phenom(tm) II X4 905e Processor 2999 MHz W=24.5 plies; 1.182kN/s B=25.4 plies; 2.097kN/s; 1.816.036 TBAs 1.d4 d5 2.c4 c6 3.Nf3 Nf6 4.Nc3 e6 5.Qb3 Nbd7 6.Bf4 dxc4 0.11/23 4:15 7.Qxc4 0.27/21 52 a6 (Nb6) 0.10/22 2:14 8.e4 (e3) 0.45/22 1:43 8...b5 0.11/23 2:29 9.Qe2 (Qd3) 0.23/22 13 9...c5 0.08/21 5:02 10.d5 0.42/22 45 c4 (Bb7) 0.01/21 3:18 11.d6 0.36/22 3:45 Bb7 (Nh5) 0.04/23 52 12.e5 0.24/21 3:10 Nd5 -0.03/23 1:16 13.Nxd5 0.06/22 8:27 Bxd5 -0.02/24 0 14.0–0–0 0.24/20 1:18 Bb7 (Nc5) -0.02/23 3:32 15.Kb1 0.30/21 2:36 Rc8 (h6) -0.06/22 8 16.h4 (Qe1) 0.55/20 2:29 16...c3 0.06/20 2:37 17.bxc3 0.48/21 8 Rxc3 0.06/21 2:21 18.Rh3 (Bd2) 0.47/22 5:33 18...Rc4 (Rc5) 0.10/20 6:23 19.Nd4 0.67/18 1:23 g6 0.20/19 1:02 20.Qb2 0.47/20 3:09 Rc5 (Rc8) 0.39/20 1:44 21.Nb3 (Rc3) 0.59/21 3:23 21...Rd5 0.37/23 4:41 22.Rxd5 0.67/24 2:09 Bxd5 0.41/23 0 23.Rc3 0.67/23 1:59 Bg7 0.40/22 51 24.Rc7 0.71/23 1:03 Be4+ (0–0) 0.49/24 14:02 25.Kc1 0.76/23 1:17 0–0 0.75/24 14:27 26.Qd4 1.28/27 3:23 Bd5 0.97/23 0 27.Bg5 1.34/24 5:31 f6 (Qe8) 1.42/22 0 28.Qa7 1.45/23 2:56 Qb8 (Qa8) 1.40/23 0 29.Qxb8 1.59/23 1:10 Nxb8 1.52/24 1:47 30.exf6 1.58/25 1:17 Bxf6 1.52/25 3:10 31.Bxf6 1.44/26 51 Rxf6 1.49/24 23 32.f3 1.45/25 32 Nc6 1.46/25 2:40 33.Nc5 1.49/26 3 e5 1.46/26 5:17 34.d7 (Nb7) 1.46/27 2:06 34...Nd8 1.25/26 2:22 35.Rc8 1.43/27 9 Rf8 1.25/27 0 36.Nxa6 1.66/26 3:25 Bxa2 1.72/27 6:12 37.Bxb5 1.53/28 1:00 Kg7 1.70/27 2:24 38.Rc5 (Kd2) 1.65/26 4 38...Kf6 1.60/25 1:43 39.Nc7 1.58/26 3 Be6 1.60/26 1:47 40.g4 1.69/26 1:38 Ke7 (h6) 1.65/26 29 41.Rxe5 1.88/25 1:55 Kd6 1.72/28 56 42.Nxe6 2.03/24 4 Nxe6 1.69/26 48 43.Re4 1.99/27 3:17 Rb8 (Nc5) 1.83/28 7:21 44.Ba4 2.09/26 3:09 Rd8 (Nd8) 1.98/27 0 45.Kd2 2.25/24 1:47 Nc5 (Rb8) 2.09/27 1:47 46.Re8 3.57/27 1:43 Kc7 2.67/29 5:32 47.Bb5 (Rxd8) 4.12/28 38 47...Nxd7 (Nb7) 4.21/27 3:46 48.Rxd8 5.30/27 1:14 Kxd8 4.03/28 52 49.Bxd7 6.19/27 8 Kxd7 4.42/31 0 50.Kd3 7.49/31 21:08 Ke7 (Kd6) #30/39 0 51.Ke3 (Ke4) 7.50/29 3:05 51...Kf7 (Ke6) #28/36 3:31 52.Kd4 7.71/29 2:46 Ke6 (Kf6) #27/38 0 53.Ke4 9.96/31 8:32 Kf6 #26/38 0 54.Kf4 (Kd5) 9.97/28 2:17 1–0

TimoK
Posts: 97
Joined: Sun Jan 03, 2010 11:28 am
Location: Hamburg

Re: Mythbuster continued ...but Komodo comes back

Post by TimoK » Sat Mar 10, 2012 12:18 am

Hello Bram,

thank you very much for your commentary on the games! That is highly appreciated because it makes the matches more interesting. For me it's not possible to watch all the games at the same time so it's good to have a second pair of eyes looking for interesting games to comment on.

Indeed Komodo is doing a good job so far especially against Stockfish (as well as Houdini also leads against Stockfish). I think all 5 engines will be very close together in the end if we look at the scores (maybe Houdini will manage to break away from the rest). But they are playing very differently, so it's kind of interesting that we have 5 different engines playing at about the same very high level. After this test I think more games will be necessary so maybe I'll add another 40 games all against each other.

Best regards
Timo

Post Reply