Re: [gnugo-devel] 261 games 3.1.31 vs 3.1.32

[Daniel Bump]

> > I'm fairly certain "against itself" means that both white and black
> > were played by GNU Go 3.1.32, and that the question is whether it
> > plays a reasonable game in autoplay.
> 
> Exactly.

Sorry, somehow I misread this.

So here's some more comments, this time for W:

28 at D3 is joseki (though a dangerous one when B responds
at E3). W needs to know how to make sabaki after crosscutting.

While B neglects the Q16 group in this game, W neglects the
bottom left group. 

At move 34 W at R15 would be severe.

38 at R15 would kill.

44: since W has two weak groups, it had better defend G4
or D12. (R12 or R15 now would also threaten the B group.)
Instead, it defends a group that is already alive.

After 45, W would like to defend the defect at Q7. There
is no time at move 46 for this, however.

This game shows that GNU Go is not good at evaluating
the weakness of groups. Of course, we know this from the
NNGS games.

Currently I'm against self-tuning because, among other
things, it diminishes the value of self play as a
yardstick of improvement.  The time spent tuning is
better spent tuning against human games, or against
other programs.

Dan