The 10th tournament Gomocup 2009 took place at April the 4th-5th in Nove Straseci at my home. There were two dual cores computers available; overclocked Intel Core 2 Duo CPU T7500 3GB RAM and Intel centrino Duo T2300e 1GB RAM. Primary schedule was to use all four cores, but there was technical problem with restraining CPU cores usage on Intel centrino computer. Because of multithreading AIs PureGM and Nabamoku, there was only one client on this computer at the beginning and later the second client was not launched, to have a possibility to solve technical problems and answer questions on the chat.
The new AI versions of Gomocup 2009 were Kanec, Hector (HFM), Hewer, Swině, PureGM a Goro. And there was quite new novice Amoeba using Monte-Carlo tree search algorithm and heuristic evaluation function tuned by the cross-entropy optimization algorithm written by Gálli Zoltán (Hungary).
There were several improvements and innovations in Gomocup 2009. The first one was the tournament system inspired by soccer/hockey league system. AI was split into 4 groups according to their last year’s result (best 6, next best 6, next best 6 and the remains + novices). The best three AI went forward from every inferior league to superior league.
There were innovations in Gomocup online: more stable (because of better hosting), including chat, new transfer technology, real-time gomoku boards with replays (written by David Hoskovec).
|Tournament||Time limit per move [s]||Time limit per match [s]||Memory limit [MB]||Board size||More than five in row wins|
Results of Gomocup 4. freestyle gomoku league
Results of Gomocup 3. freestyle gomoku league
Results of Gomocup 2. freestyle gomoku league
Results of Gomocup 1. freestyle gomoku league
There were 3 AIs went forward into 1st freestyle gomoku league: Pela, new Kanec and new Goro.
In the first round, Goro wes excelent and lost only one match.
In the second and the third round was Goro drawing up by Onix, HGarden and Hewer.
The most exciting part of tournament, unfortunaly it was after midnight.
At first was played match Onix versus Goro, which could estabilish the winner, but it was draw and total score was 3:3.
There was probably win missed by Onix in deep about 22 moves ahead.
From that time it was cleat, that Goro can keep leadership only by losting no more than 5 matches and one point.
About 4 matches before end of the tournament Hewer was still striking for winnership, but lost important match with swine.
Not until the entirely last of tournament match Hewer versus Goro resolved the Gomocup winner.
If hewer wins both of two matches, the winner will be Goro or Onix (depending on number of moves and time of thinking), otherwise Goro.
The first match was win very quickly by Hewer. And the second match was very exciting,
Hewer forced Goro to get into jackpot and Hewer reported (by debug message) Hewer’s win. It looked like,
there will be two winners of Gomocup (Goro and Onix),
but Hewer failed in clear situation and due to bug missed the win and Goro win the match and Gomocup.
I analyzed this situation, the bad move is move number 14(indexed from 0) replay freestyle1/8×7-32(1).psq.
The right move in this situation was [9,11] probably followed by win in little more then 40 moves.
This Hewer’s bug bereaved Onix 8 seconds before end of tournament of possibility to be a winner (depending on number of moves and time of thinking).
The deepest reported and founded wins were reported by Onix (28 moves ahead) and Tito (28 moves ahead) and Hewer (38 moves ahead, freestyle1\0x8-62(1).psq).
In the last year swine reported and found win 46 moves ahead, the new swine “only” 26 moves ahead.
There was launched standard gomoku tournament for the first time in gomocup history.
So far, this rule was supported only by Pela and from now it is also supported by artificial intelligence Hewer form version 3.1.
Standard gomoku AIs are more desired by advance and profi gomoku players.
Perhaps estabilishing of this category will make Gomocup to be more useful for these gomoku players.
There is a simle test for AI to enter this category. There is a situation in which AI can make six in row or defend the three. Only Pela and Hewer was defending the three.
The winner of this category was Hewer.
There were selected almost the same AI to the fast-game as in the last year,
Onix was removed due to 74 overtimes in the last year and new Amoeba was added.
Because of HfM 2009 was not tuned for fast-game, I used Hfm 2008, but I didn’t notice,
that it is expired. In that time I obtained new HfM 2009 fixed for fastgame.
That is way there was one another game HfM versus all and the results was merged.
The winner of fastgame was Hewer followed by HGarden and Kanec.
You can download complete results and openings here