[Computer-go] John got tromped by Zen 1-3

[Darren Cook]

> Basically we have concluded based on a mere 3 games that John Tromp was
> "crushed" by the computer as if there was no hope had the match continued.

If you treat it as just 4 samples, that is a reasonable point of view,
statistically.

But you can gain a lot more information by reading the comments by
stronger players and counting the number of notable mistakes by each
side. This gives you a data sample over the 4 games as something like
20-15 for John-Zen. (No, I didn't count them, it was just my impression.)

Still not enough (*) to claim Zen is stronger than John at 90%
confidence, but it is a stronger claim. (At the expense of being
subjective ;-)

I just wanted to point out how an infrequent, well-publicized event can
generate quite a lot of useful information by attracting a strong audience.

In fact, getting on my favourite soap box, I'd say the quality of
information generated is infinitely more useful. In a tournament of 30
games, coming first, when you came third a year ago, tells you: "You're
clever. Something you've done in the past year was also clever."

Being shown a list of 15 mistakes by players who are a few ranks
stronger gives you 15 specific examples of where you need to improve if
you want to improve two ranks.

Darren

*: I think my count needed to be 22-13 for 90% confidence.

> prop.test(22,35,alternative="greater")

	1-sample proportions test with continuity correction

data:  22 out of 35, null probability 0.5
X-squared = 1.8286, df = 1, p-value = 0.08815
alternative hypothesis: true p is greater than 0.5
95 percent confidence interval:
 0.475438 1.000000
sample estimates:
        p
0.6285714



-- 
Darren Cook, Software Researcher/Developer

http://dcook.org/work/ (About me and my work)
http://dcook.org/blogs.html (My blogs and articles)