[gnugo-devel] Performance Increase by caching ladders and patterns

[Chris Smith]

Sometimes in a game a situation arises where a standard pattern is 
expected, such as a getta pattern or a ladder.  The pattern 
matcher/ladder code is going to make sure that there are no ladder 
breakers/extra liberties before GnuGo pursues- however this is 
calculated after each move (correct?)

Would it be a good idea to simply have a 'scratchpad' of sorts to 
remember these situations on the board and earmark the points involved 
so that if say the other player lays down a ladder break GnuGo will 
instantly know that that will break ladder X and then evaluate whether 
to kill or let that one escape.

Ladders however are just an example, this could be extremely useful is 
where GnuGo reads ahead and makes some killer move to kill a set of 
stones (actually just matches some pattern).  But could it 'remember' 
what pattern it was- and some of the subsequent variations, followups 
and/or what it thought about to arrive to that move?

This idea however should be debated because we wouldn't want to simply 
'respond' to a move because it reminds GnuGo of something 50 moves ago, 
however it can be used to quickly give GnuGo an idea of its opponent is 
trying to do.  Such as using the influence of one group to force a 
capture.

I know this seems a little abstract, however I am playing a game where 
a group in the corner has no chance of making eyes- however GnuGo seems 
to be taking almost two minutes per move when it probably has already 
read through all the possibilities.

Anyways, I was just wondering what your thoughts were on the subject.

-Chris Smith