Re: [gnugo-devel] sente analysis

[Paul Pogonyshev]

Kevin Yong wrote:
> Hi:
> [sente/gote analysis suggestions]
> [...]
>
> Any comments/suggestions are very welcome.

this is certainly an important issue, but it doesn't look like being
high priority at the moment for several reasons:
  - we are simply not ready for it (just like you wrote) and such
    followup analysis would require major code changes
  - there are more important yet probably simplier problems to solve
  - the problem can be alleviated with existing tools, gnu go is
    already decently good at determining followups
  - to be really complete, such analysis would have to be several
    moves depth in reading. e.g. consider these positions:

OOOOXXXX
...OX...
...OX...
....*...
--------

here, the marked move is indeed sente, but to determine it by reading
we need to analyze four moves.

OOOO...
OXX*...
OO.XXXX
..OX...
..OX...
-------

(maybe not the best example, but) here the marked move is in effect
gote: also it threatens to capture black stones, black's answer is a
threat in turn. here we need to analyze three moves and there are
certainly positions that require more than three moves for accurate
analysis.

so, such reading might be too slow and there are probably better
ways to spend cpu time.


as a general comment (please don't take as an offence):

there are quite a few of us working on gnu go at the moment and i'm
pretty sure we all know its major weaknesses. so far your suggestions
have given almost nothing (apart from some fixed tests as i remember).
it would be helpful if you started actually _coding_ something. it can
be a far smaller improvement than you aimed at in this and other mails.
every single unexpected pass in regression suite is an improvement
(unless it comes with a couple of fails of course ;)

Paul