Re: [computer-go] Pattern Matcher

["Frank de Groot" <frank@xxxxxxxxxxxxx>]

> Sorry Frank, but I think a pattern-matcher is not in the same class as a
> full life-and-death program, even if it has the limitations that Wolf's
> program seems to have.

It's not really a plain pattern matcher.
It comes with the patterns.
I charge 1 dollarcent for a pattern, 8,388,608 patterns = 100,000 Euro.

The resulting system is able to predict pro "Joseki" far beyond move #100:

http://www.moyogo.com/popo_gomonster.htm

In fact it's not Joseki any more, as long as a cluster of stones has ever
been on the board in half a million strong games, my matcher will see it.
This is incredibly powerful. Basically, I have solved for Go what chess
programmers took many years to solve for Chess (huge opening libraries
classified into relative "goodness").

It contains all patterns that occur more than once per 100,000 strong games.
I can say "all" because I catch 8 size classes, from tiny to entire board.

The whole "positional" and "strategical" aspect of computer Go is basically
"solved" by this pattern system so I can dedicate myself to tactical search.

The pattern matcher itself is a jewel but the value lies in the patterns.
I think it's a price hard to beat, a cent for patterns like shown in the
URL.
Almost ALL of those 8 million patterns are patterns like that.
Good luck typing them into your "much more suitable" pattern matcher
system..

And, like I said, harvesting 8 million patterns is not even the biggest
deal.
The real value (included in that 1 cent per pattern) is a lot of statistics.
That 1-cent pattern comes with a frequency and a whole-board correlated
value based on statistical analysis of half a million games.

So, 100,000 USD is cheap and yes, I think it is of the same relevance and
usefulness (probably even much more) than the GoTools source. SmartGo has
GoTools incorporated and it is a low-ranking Go program.
Any Go program that would incorporate my pattern expert system would be able
to play along with pro-level further-then-move #100 "after-Joseki". And it
makes the correct decision as to which moves are more relevant than others
(it will take care of pressing issues and then return to a Joseki etc.) No
current systems are able or will be able to do that.

Now, my patterns are not simple matching stones. They also include how many
stones the chains have and how many liberties they have and they include
edge info. Which means that especially the smaller patterns pack more punch
than you would think. They know what's beyond them, so to speak.


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/