Re: [Please Help] Pattern Matching Comparison

[Mousheng Xu <xu@xxxxxxxxxx>]

At 11:27 AM 3/12/99 -0600, rbrown@xxxxxxxxxxxxxxxxx wrote:
>> If you use 0 for empty, 1 for black, 2 for white, 3 for
>> jie, then you can use 2 bits to represent one poisiton on the board.
>  ^^^
>
>I must ask; what is jie?  I don't recall ever having seen or having heard
this

	Hey, that's a good catch. :) "Ko" is the Janpanese way to say "Jie".

>One of the really difficult questions to answer if you want to use patterns
>is this:  Just exactly what sort of patterns do you want to store, compare,
>recognize, and manipulate?

	If the move is good for protecting itself and/or killing the enemy.
Fuseki, joseki all fall into the same category. I need a single pattern base.
	
>
>Another difficult question (perhaps just a rephrasing of the above):  What
makes
>one pattern, or more precisely one pattern _classifier_ "better" than
another?

	For simplicity, there is just one information to keep: number of enemy
stones it can kill after a certain number of moves.
	So you guys have not discussed my original questions yet: 1. pros & cons
of PAT, Zobrist, Silly Number, Pattern tree; 2. which do you prefer if you
have to choose one? why? 3. Any other suggetions about patterns?

>Here I must disagree.  Patterns of _behavior_ in go are far _more_ important,
>not "almost" as important to the programmer than the spatial patterns are.

	Great new word -- "behavior!" Don't you just mean play history? 

>One very interesting thing about constructing a pattern classifier -- and
this
>is true whether the patterns we want to classify are medical histories,
weather

	I just told you :-) : the single pattern classifier is how many stones it
kills.
>
>For example, consider a medical diagnostic test which uses patient data
(blood
>sugar level, white cell count, body temperature and so on) to classify a
patient
>as either having a certain disease or not.  Let's say that this test errs
>occasionally, and suggests that some patient has the disease even though he
>really doesn't.  On the other hand, it rarely, if ever, fails to identify
>someone who does in fact have the disease.  Now just because there are a few
>"false positives" from this test, we don't throw the test out as useless.

	We all use heuristic algorithms, and that's what I am trying to avoid at
this beginning state. I want to be a perfecist first, and then surrender to
inperfectism. :)

	Boy, you are good. You have read so many books. With that much knowledge,
IT's A CRIME not to help me to resolve my multiple choice question for
patterns. :)

	Thanks a lot.

-- Mousheng Xu