Re: computer-go: Go4++: Passing in tsume searches.

[Dave Dyer <ddyer@xxxxxxxxxxx>]

>
>Am I the only one who's searches are clogged up with 
>passes? How have other people dealt with this?

I had the same sort of problem.

There are really three kinds of passes in go.

1) Mandatory passes due to Ko (or SuperKo). 

In Tsumego, you do not need to separately analyze branches
in which a Ko is repeatedly captured and recaptured. Recapturing
an open Ko should be forbidden throughout the subsequent tree.
The value of the position in which the Ko is recaptured is implicit
in the value of all the siblings of the move the opened the Ko
in the first place.


2) Optional passes where you trade the tenuki for locally reduced value.

In Tsumego, you never need to consider these moves.  In the full
game, you sometimes explicitly construct a situration where
one player has moved twice in a row at top level, to see if the group's
status and territory value is different, but within local searches
you don't need to consider moves that result in a less than best value.


3) Mandatory passes because the position is Dead, or making another
move would reduce the qualitative value of the position (perhaps by 
converting seki to dead, or alive to seki).

These are the ones that are hardest to deal with.  However, they always
occur in situations with very restricted liberty count, and therefore very
small numbers of potential moves.  As a heuristic, you can consider
passing only if the liberty count (or number of possible moves) is
less than N.  Unfortunately, as the examples brought up recently have
shown, there is no correct upper bound on N.  I believe you can construct
postions with any nuber of liberties where a pass is mandatory.  So, you
will occasionally misanalyze a position.  The upside is that your actual
outcome will be better than you expect - When analyzing the
problem, if you must pass, and you do move, and as a result kill yourself -
you will decide not to move and possibly will end with a live group.