Re: [Please Help] Pattern Matching Comparison

[Mousheng Xu <xu@xxxxxxxxxx>]

At 06:33 PM 3/12/99, Darren Cook wrote:
>>    1. PAT Tree: guess all of you guys know what it is. If not, go read some
>>algorithm books.
>
>What is a PAT Tree? :-)
>
>Yes, I looked in an algorithm book and it doesn't have it. Closest was
>Patricia Tree - is that what you mean?
>

	Yes, Sir. "A PAT tree is a Patricia tree (Morrison 1968; Knuth 1973;
Flajolet and Sedgewick 1986; and Gonnet 1988) constructed over all the
possible sistrings of a text".


>
>A few thoughts about how humans use patterns (and IMHO how computers should
>use them):
>
>1.Stones and points have meaning. If your patterns just store
>empty/black/white the program won't appreciate the real meaning. The larger
>the pattern the less this matters, but the fewer times your pattern can be
>used. A database of whole board fuseki patterns is an example.

	Sure, you could and should associate with each pattern a lot of
information such as the potential value, connection strength, good for task
A, bad for task B, etc. 
	I don't understand the part "the larger the pattern the less this
matters". What is "this"? 

>
>2.Variable size and shape. Patterns are normally bounded by features. A

	This is a good idea: pattern size & pattern function are related. But this
worries me, a unifier, a bit. There is one goal: to survive (and/or to
kill), whether you want a fuseki, yose, or tesuji. Why do we cut the goal
into pieces? :) I will think it over.

>3.The pattern accross time (a.k.a. moves sequences; joseki) is almost as
>important as the spatial pattern [1]. [2]

	Sure, and that's why a pattern tree is good. But the problem of a pattern
tree is sub pattern match.

	Thanks a lot for your discussion, Darren. It's really enlighting.

-- Mousheng Xu