Twenty-first KGS Computer Go Tournament

Sunday December 3rd, 2006

These results also appear on official KGS pages: Formal Division, Open Division which link to the game records.


 Formal divisionOpen division
board size19x1919x19
time28 minutes absolute28 minutes absolute


Formal division: five-round Swiss.
Open division: five-round Swiss.


The first round started at 09:00 UCT for the Formal and 09:10 for the Open division.


As usual, the tournament was held in two divisions, Formal and Open, with more restrictive entry conditions for the Formal division.

Formal Division   19x19


Open Division   19x19


The "real" names of the bots listed above, and of their programmers, are listed here: programs which have registered for KGS Computer Go Tournaments.


Mango had registered for the Formal division, and housebot and HbotSVN for the Open division, with their owners warning me that they would be playing unattended. At 21:40 UCT the evening before the tournament started, the KGS server went down. When it restrated, these three bots were no longer connected. They lost their first-round games by forfeit, and I them withdrew them from the tournament.

Formal division

In round 1, valkyria was drawn as white against the absent Mango. When it did nothing for ten minutes or so (while waiting for Mango to connect and move), the server kicked it out of the game for idleness.
      This must be wrong. A player which has done nothing wrong should not be penalised for a lapse by its opponent. And it is a serious consequence of the way the server works: it could lead to the win being awarded to a player that turns up late, if its opponent's operator does not spot the problem, kill it, and relaunch it. I hope that wms will change this behaviour.

In round 2, GNU had connection problems in its game against AyaBot, and eventually lost on time.

In round 3, GNU's connection problems continued, allowing MoGoBot a win on time.

In round 4, AyaBot played more slowly than usual in the endgame, and lost to firstgo.

Open division

In round 2, MoGoBot19 had an easy win in a finished game against WeakBot50k, with 35 seconds left on its clock. But instead of moving or passing, it allowed its time to run out.

In round 4, MoGoBot19 had another easy win in a finished game IdiotBot. This time it continued to play, needlessly filling its own liberties while IdiotBot passed, until it had again lost on time.

As far as I know, MoGoBot and MoGoBot19 are the same program. Yet MoGoBot won 3/5 in the Formal division, while MoGoBot19 won only 1/5 in the Open division, against weaker opposition. My guess is that, being a Monte Carlo program, it finds time a problem when playing on a full board, and has been tuned so as to use almost all of its time allocation. So against programs that play "sensibly", and particularly against the well-mannered valkyria that resigns when it has a lost position, MoGo can win; but against programs that end with many dead stones on the board, MoGo can lose on time while trying to tidy up a won position.
      Sylvain Gelly has explained: 'Yes it is the same program as MoGoBot, running in a different machine. It turns out that someone else was using the machine were MoGoBot19 was running, and as we use the "CPU time" and not "real time", MoGo was taking a lot more real time for its moves than it should. It is why it lost games on time, and some playing very badly at the end (when time is < 20 s, then MoGo plays randomly, it is a "security"). So for MC programs time is a "problem", but it is not the case here, the problem was "external" ;). As I was in a plane I could not operate it (a friend was there if there was a crash, but he could not know about the machine).
      I am very sorry for the observers about the bad games this problem gave.'