[Computer-go] Beta-testing: feedback to bot owners

[Nick Wedd]

I first thought of using 7 komi for kgs 9x9 bot tournaments during a 
discussion with Olivier Teytaud.  He is planning to hold the 13x13 and 
19x19 Computer Go events, in the EGC in Bordeaux this summer, on KGS, 
using its tournament scheduler.  He was not planning to use it for 9x9 
because, as he wrote, "the random pairing in 9x9 is a disaster; it's 
clear that having white is better in 9x9 Go with komi 7.5, so the games 
might become a coin toss".

I suggested that if 7.5 is too much, he could use 7.  I decided to test 
this, and offer the bots some practice, by running some 9x9 events with 
7 komi on KGS.  But two things have happened to make this irrelevant: 
Olivier has told me that he is not willing to use 7 komi, and the 
disastrous random pairing has been fixed - or will have been, when the 
current beta-test version of KGS is installed in place of the current 
version.

The 9x9 beta-test tournament that I ran on Thursday shows what is likely 
to happen if one bot understands integer komi while the other does not. 
GNU Go version 3.7 and Orego12 played each other six times.  Orego is 
clearly the better player.  The three times that Orego was black, it won 
by 2 points.  The three times that Orego was white, the result was jigo.

However, several people have expressed interest in a 9x9 event with 
integer komi.  If a majority of likely entrants is in favour, the next 
9x9 KGS monthly bot tournament, on April 3rd, will use a komi of 7. 
Please send your votes "for integer komi of 7" or "against integer komi 
of 7" to me at maproom @ gmail dot com.  I will treat the votes as 
confidential.

Nick
-- 
Nick Wedd    nick at maproom.co.uk