[computer-go] Re: Amsterdam 2007 paper

[Sylvain Gelly]

Hi John,

> You mean they are positions in which humans passed and scored the game?
> And you're just testing if the simulation correctly converts all
> territories into 1-point eyes,
> and kills all invasions?

Actually MoGo (10k simulations) labelled the positions, then a human
checked the labels. The positions were labelled if and only if MoGo
thinks the winner is sure at 80%. So it is why it is rather end game
positions, but not completely end game.

You can have a look at the positions in sgf here (but those sgf does
not give who is to play unfortunately. The native database is not in
sgf):
http://www.lri.fr/~gelly/mogo/positions/

Maybe there are some mistakes in the labels, but if any they are few
and the result holds (we've go strong evidences).

Sylvain