[gnugo-devel] Pattern A1120

[Portela Fernand]

Hi,

This is just some news and thoughts about my work on this pattern.

I spent a very long time trying to modify it and achieved nothing very
useful. Yesterday I just tried to remove it and work from scratch. I've
started adding some patterns and as of now, I have a performance boost of
a bit more than 20% (300K owl nodes saved) and a regression breakage of
23 PASSes and 23 FAILs. I still have a lot of work left, but it definitely
looks promising.

Still, this study showed me a couple issues, which will be difficult to
ignore. Maybe what follows is long time known here, sorry about it if it's
the case.

The first issue I've been facing is the fact that the escape map is static
and doesn't get updated during owl reading. At stackp >= 10, things might
have changed a bit around the attacked dragon and there's a good number of
patterns which rely on owl_escape_value()... I've made small attempts at
solving this, but nothing useful so far. Anyway, it doesn't seem to be
very urgent, although I think it should be addressed soon or later.

The other issue is the important assumption that a dragon is an
unsplitable unit. Although I can understand the reasons which pushed to
that choice at first, it seems simply unacceptable to me. A couple
examples: I've seen the engine play 3 successive one space jumps and
declare WIN by escape while all the original stones were getting cut (and
killed of course). Another time I saw the engine declare WIN by "2 or more
secure eyes", but it hadn't noticed the 2 eyes just got disconnected !

I have a couple ideas about the 2nd issue, but I wish to get some opinons
about it before I start working. So, what do you think ?

/nando