Re: [gnugo-devel] tuning patch

[Gunnar Farneback]

Arend wrote:
> I think one should add a line
> 
> >return (!xplay_attack(b,a))
> 
> (with b at ! and a at Q) to the pattern. Otherwise 'a' might get played
> because of a followup value, although the lost liberty will mean we have
> to come back and defend after the 'a'-'b' exchange.

Sounds reasonable.

> Also, maybe it is more convenient to have "Intrusion6" generalized
> as below?
> 
> 
>  ?QX
>  X!!
> -x..
> +x.x
> 
>  :8,B,value(30)

That's probably okay.

> I wondered whether it might be useful to generally add intrusion sources
> around 'dead' opponent stones (where 'dead' means either dead or
> critical and of the color not allowed to make the next move).
> 
> At the moment positions with a dead string in a chain of stones
> surrounding territory are valued poorly by the influence code.

I'm not sure. I'm afraid that wouldn't work too well in some
positions. (Sorry, no specifics, this is just a feeling I have.)

> I have worried about that shape, too! In another test case (where * was
> on the 2nd line and could get captured directly after the X at 'd', O at
> 'c' exchange) I didn't understand why it didn't get rejecected by the
> atari_atari blunder test, even if I set allowed_blunder_size to 1
> (sorry, can't find the test case at the moment).

I think the problem in that case is that *_confirm_safety() doesn't
care whether the move itself is safe.

> Can you explain what atari_atari_confirm_safety actually checks (or
> should check)?

It tries to detect whether a combination attack sequence has appeared
that wasn't around before the move. Certain stones are disregarded, as
decided by compute_aa_status().

/Gunnar