[computer-go] Learning and Go

["Robin Kramer" <robin@xxxxxxxx>]

Hello all,

I have written a neural Go program(in matlab) for a class, unfortunately I
do not have statistical results other than it beat a random player 19X19.

I am planning a decision tree program to "Learn" to play go.  My reasoning
for choosing decision trees, being that information lead effects are very
important in the opening, decision trees are fast to evaluate, and can
eliminate irrelavant data.

I also have the personal goal of learning to play go, so I would be
interested in learning the results of the decision tree even if it does
not play go well.  I think one of the interesting facts in the world of
computer chess, is that the professionals started using computers to
train, even before computers were winning.

The first question that I had was if I should take advantage of symmetry,
which is important when choosing data structures while parsing files. 
Does any one know of freely available C sgf libraries?

While with a 9x9 board there is a problem with insufficient data, which
most learning algorithms are specialized for, the 19x19 board has much
more data available, in fact maybe too much data.  I think I would choose
to train my algorithm on an unseen example rather than a rotated version
of another.  By the way FFT's are not rotation invariant.

(Assuming Laziness is a virtue) Any suggestions, pointers to file parsing
code(in C/C++, not matlab), or information about decision tree
architectures which have already been attempted would be appreciated. 
Thank you all in advance!

Sincerely,

Robin Kramer


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