This is not an official KGS page.

Eleventh KGS Computer Go Tournament

Sunday February 5th, 2006

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

Rules

Formal division:
13x13 board
Chinese rules, komi 7½
18 minutes each absolute time
Open division:
9x9 board
Chinese rules, komi 7½
13 minutes each absolute time

Format

Both Swiss

Times

The first round started at 10:00 GMT for the Formal and 10:05 for the Open division, subsequent rounds started at forty-minute intervals and thirty-minute intervals respectively.

Results

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

Formal Division   13x13

placenamewinsSOSSoDOS
1stGNU61212
2ndviking54164
3rdCrazyStone2200
4thfirstgo0240

Open Division   9x9

placenamewinsSOSSoDOS
1stbotnoid73630
2ndtwinbot64028
3rdStoneCrazy63118
4thMango53214
5thtlsBot53112
6thantbot9x94317
7thSimpleBot3314
8thWeakBot50k3251
9thHouseBot1330
10thIdiotBot0300

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

Description

Fourteen games ended in arguments between the bots. Records of all these "forfeited" games are now available below.

Formal division

The results table is very clear. GNU won all its games, beating each opponent twice. Viking5 beat each opponent except GNU twice. CrazyStone beat firstgo twice.

The event was scheduled as a Swiss, not as a double round-robin; so the draw algorithm did not arrange that, the second time two bots played, their colours were reversed. I do not see this as a problem.

Open division

After seven rounds, twinbot and botnoid were tied for first place on six points each. They had already played each other (in round 3, twinbot winning), so it looked as if they might each end the tournament with the same number of wins. However the onlookers were surprised, and pleased, to see that despite their having played each other before, they were drawn against each other again for the final round. Botnoid won this game, and the tournament.

Cat and Mouse

Twinbot's game against WeakBot50k in round 6 was entertaining. Twinbot is a GNU Go derivative, so it was no surprise that it beat WeakBot50k easily. What was entertaining was the way that twinbot appeared to play with its opponent, like a cat with a mouse.

Twinbot's first pass was on move 64, with all the black stones dead. Twinbot's move 92 was a pass that allowed Black to make an eye in its largest group. With move 99, Black filled this eye. Twinbot's move 214 was a pass that allowed Black to connect two groups, making an eye. With move 231, Black filled this eye. Twinbot's move 236 was a pass that allowed Black to make a new eye. Twinbot later captured the group. Twinbot's move 276 was a pass that allowed Black to make a new eye. Black finally passed with move 299, because it had no other legal move; it then conceded that its three groups were all dead. Twinbot won by 88½ points; this would have been 190½ if Japanese rules had been in use, quite a margin for a 9x9 board.

This makes it sound like time-wasting, but in fact little time was used by either program. The last 200 moves of this game took exactly a minute.

Scoring Problem

The game Mango-antbot9x9 in the final round presented a problem. Both players passed in the position shown, and it was not clear how the postion should be counted. Several views were expressed:

  1. With anything near to reasonable play, the right half of the board is all seki, and White is going to play A7 or A8, so White is more than 12 points ahead on the board, plus komi.
  2. There are unfilled dame, so nothing can be counted as territory. White wins by the komi. (I believe that this is the result is consistent with a strict application of Japanese rules.)
  3. Any territory in the seki cancels out, but White actually owns D3. So White is one point ahead on the board, plus komi.
  4. W has 31 stones on the board, plus D3, plus one of J7 & J9. Black has 29 stones on the board. So White is four points ahead on the board, plus komi.
Fortunately, all of these agreed that the game is a win for White.

However, such a position could easily give a problem to the organiser. I hope I never have to deal with such a problem. There are two ways of avoiding it:

  1. Get the KGS game-end protocol working, and require the players to support it.
  2. Persuade the players that if your opponent has some stones which you believe are dead, and it is conceivable that your opponent does not realise they are dead, then don't pass until you have taken them off the board.
Of course, these two ways are compatible. But the latter applies even when you are competing in some other event, which does not use kgsgtp.

Forfeited games

Here is a list of the "forfeited" games. You can download any game listed here by clicking on it. Other game records are available from the KGS pages: Formal Division. Open Division.

roundFormal divisionOpen division
1 botnoid-Mango
HouseBot-antbot9x9
2CrazyStone-firstgoantbot9x9-botnoid.sgf
HouseBot-tlsBot
3 botnoid-twinbot
HouseBot-SimpleBot
4  
5 SimpleBot-botnoid
twinbot-tlsBot
WeakBot50k-HouseBot
6 botnoid-tlsBot
Mango-HouseBot
7  
8 Mango-antbot9x9
twinbot-botnoid