Re: [computer-go] Pattern Matcher

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 14:26 7-11-2004 +0100, Frank de Groot wrote:
>Hi Vincent.
>
>Please don't overquote, please read my posts carefully before asking
>questions I have already answered and, before all else, don't waste my time
>saying silly things or engage in religious fundamentalism.

Please don't act overarrogant. I have met past 15 years enough arrogant
computer game researchers to recognize one when i see one. In fact i'm sure
without arrogance and stubbornness one can't write a program of his own.

But please don't do as if i know nothing from patterns. I work fulltime on
patterns. Not so much in go, but in chess.

I happen to know exponential more than you there about the advantages and
disadvantages of patterns.

Note that my religion is Christian. My religion isn't condemning automatic
generated patterns used either for move selection or evaluation.

In your completely wrong and naive calculation below you forget 1 important
thing which all the automatic approaches on games seem to forget. 

That is that those 500000 games you have, they are not played by 1
professional player on a board for his own at which he demonstrates to
beginners which patterns are good to accomplish, thereby playing complete
idiot moves for the opponent.

There is 2 players who both very well known what is good for them and what
is bad for them. 

So a few things do not happen in their games:
  a) the real bad patterns do not get on the board,
     because a professional player always will manage to escape
     from the real big idiocies, such as what happened to me when i
     lost with 361 points difference against a strong go player. 
     And that very instructive game is *not* in your database.
  b) the ultimate goal of a certain construction never gets shape on the
     board. Usually 'compromises' get on the board.

So from a&b we can conclude that you only can recognize 'compromise'
patterns from the database. 500k games isn't enough then.

The additional problems you have:

a) the *size* of 1 pattern to automatically recognize. there is
   10^100 possibilities for patterns so there is many possibilities
   for an automatic recognizer to consider size and shaping of a pattern.
b) probabilistic life&death evaluations from human estimation. A human
   will simply consider the statistical chance that a certain group 
   survives, expands or limits the opponent. That influences the 
   decision elsewhere on the board to take a decision on laying 
   down that stone. If you take such human factors in a computer model into
   account the number of possibilities for a single pattern in database 
   soon grows to several trillian instantly.
c) the speed at which an automatic pattern recognizer can search.   
   handcoded knowledge has a major speed advantage.
d) no one wants to pay 1 cent for an automatic pattern. 
   The price for professional software is around $5-$20 for each effective
   well code line of source code. That's for non GUI code :)

>For the rest I thank you for your interesting contributions here.
>
>> Your problem is the statistical significance of a pattern.
>
>Correct.
>
>> 8 millionpatterns would require like 8 billion professional go-games to
>even have a
>> slight proof that a certain pattern works.
>
>Incorrect on several counts.
>First of all, your claim is a religious claim.
>I on the contrary am the one doing the actual research, and I found that
>when you take 125 billion patterns from half a million games, there are
>about 1 million patterns that are very frequent (occur thousands of times).
>Then there are about 3 to 4 million patterns that are less frequent (I end
>up with a few dozen of them) and about 4 million patterns that occur once or
>twice per 100,000 games. All the rest is rarer than that.
>
>Now, the proof is in the pudding, the pudding being the ability to predict
>moves in never-before-seen pro games.
>When you say that it can't work but I see my system predicting long
>sequences in pro games that are played after I trained my system, I see it
>in fact does work.
>
>> How many professional go-games do you have in your database?
>
>I trained it with around 465,000 games, all high ama or pro.
>
>I already mentioned in your quoted post that my database contains 52,000 pro
>games.
>But I discovered that ama games on certain Go servers can be of much higher
>quality than (older) pro games.
>I prefer to keep a secret which ama games that are, and why.
>
>_______________________________________________
>computer-go mailing list
>computer-go@xxxxxxxxxxxxxxxxx
>http://www.computer-go.org/mailman/listinfo/computer-go/
>
>
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/