Re: [gnugo-devel] moyo reduction

[Gunnar Farneback]

Arend wrote:
> My general assumption is that moyo reduction moves should be treated
> in the same way as other moves as far as this is possible. In particular,
> GNU Go should be able to evaluate their territorial effect reasonably
> with its usual valuation procedure. If this is successful, we could on
> the way also drop the enforced minimum value of such patterns as LE14,
> LE6, LE11.

Agreed. This mirrors my thinking.

> 1. The current territorial valuation with its 3 steps (area = 0.13,
> moyo = 0.35, territory = 1.0) is too coarse and should be replaced
> by a continuos function interpolating between the existing
> threshholds.

Yes. The reason for the three steps is mostly historical, since that
was what the moyo algorithm in 2.6 provided.

However, one should be aware that the influence function is used for
multiple purposes, which probably should be less tightly coupled than
they are today. Except for territorial evaluation, the results of the
influence function are also used to compute escape values and through
autohelpers in various pattern databases, primarily conn.db and
patterns.db. Furthermore the surrounding moyo sizes are used in the
estimation of dragon safeties.

Mostly the influence function seems to be doing a reasonable job in
the other functions, so one should take some care not to break those.
It's probably a good idea not to make any changes to the definitions
of area, moyo, and territory in terms of influence values. Instead the
territorial valuation should compute it's "level of territory" measure
directly from the influence values. This has been something of a long
term plan and my most recent changes to the delta_territory
computations were made with this in mind.

> 2. Moyo reduction moves typically have a big followup moves that cannot
> be easily valued by patterns; the same is true for every move that sets out
> into a wide open space. I suggest a simple workaround for this: In
> compute_move_influence, the move played could get a strenght of s.th.
> around 150 instead of 100. Again this is a delicate tuning matter.

Might work, but I have a feeling it isn't effective enough. Another
trick to consider is to modify the attenuation value to get a farther
reaching influence source. You could also add additional influence
sources say two steps away in open directions.

> Of course, this does not find moves that break into opp's moyo by
> identifying weaknesses in the opp's connections around it, and then
> using techniques as cross-cut, sacrifices etc. Go4++ seems to do s.th.
> like this, but I don't think GNU Go could learn a careful application of
> such moves without some bigger changes in the engine.

Don't say that. I think this ties in nicely with connection reading,
which I understand is pretty much the approach Go4++ is using.

> So, once again I add more to the to-do-list than to the actual code :-),
> but I feel these ideas could benefit from some further deliberation
> while I am still working on the issues I raised with the strategic
> effect. I would be happy about any comment!

This is worth experimenting with. It's not on the top of my personal
agenda but I'll give any help I can.

/Gunnar