Google
 

Darren's Computer Go Pages: Strength Correspondance 9 And 19

It is often thought that 9x9 go is just a game for learning the rules, and contains no challenge for players of intermediate strength (e.g. 5 kyu) and above. However if that was truly the case then the win rate between a 5 kyu amateru player and a 9-dan professional would be 50-50. If you were to set up the match you would soon discover it would be 100% to the pro. Furthermore it turns out the game mirrors 19x19 strength right up to the highest pro levels; excepting players who have done a significant amount of 9x9 opening theory study.

Exhibit One

Most online servers do not rank 9x9 games, so it is difficult to get ready-made statistics. Little Golem currently combines the 9x9, 13x13 and 19x19 ranks, which muddies the waters a little, but still allows us to see some trend, as we look at each division.


http://www.littlegolem.net/jsp/tournament/tournament.jsp?trnid=go9.ch.21.1.1
http://www.littlegolem.net/jsp/tournament/tournament.jsp?trnid=go9.ch.21.2.1
http://www.littlegolem.net/jsp/tournament/tournament.jsp?trnid=go9.ch.21.3.1
http://www.littlegolem.net/jsp/tournament/tournament.jsp?trnid=go9.ch.21.4.1
http://www.littlegolem.net/jsp/tournament/tournament.jsp?trnid=go9.ch.21.5.1

We see a general progression of higher-ranked players beating lower-ranked players. Also note how the wins are not scattered; these charts generally show a new perfect triangle, with 7-1 or 8-0 scores for the top player.

OGS is another server that appears to merge 9x9 and 19x9 games to make the ranking. This URL shows the qualifying rounds for a 9x9 tournament.
http://www.online-go.com/tournaments/main_viewround.php?tID=26&cID=2

The same general trend is visible, when skimming through all 22 groups. (There are exceptions, and the triangle is not clear in some groups, but in most groups it is.)

Exhibit Two

The minigo TV series played blitz 9x9 games between two pros, over a number of years. The winning pro stayed on the board, and played another pro. The lower-ranked pro very rarely won, and if they did were even less likely to repeat the trick the following week. The 7-9 dan consistently beating the 1-3 dan was an obvious trend. The less obvious one was the 8-9 dan beating the 6-7 dan [TODO: this observation may not have any statistical signficance, so take it with a grain of salt until I get chance to analyze the data properly.]

TODO: analyze the available game records to get a win rate by pro rank.

Note: the blitz nature of the games will introduce noise, but may also have had some effect on the outcome; at longer thinking time the differences between pro ranks may disappear. Unfortunately pros never play 9x9 at long thinking times as far as I am aware, so we can only guess what would happen.


© Copyright 2007 Darren Cook <darren@dcook.org>