5 in a row Tic Tac Toe

Thanks for the reply, i've search already the work of Victor Allis,
but all i get is broken links. In my case i'm not really interested in
a different search, the main objective of my work is to develop good
heuristcs to the minimax wtih alpha beta pruning, and heuristics to
prune irrelevant states of the board, eg. the states in the corners are
irrevelant and therefore i will not generate them, this way i can
reduce the space and time complexity, i'm trying to discuss things
like this.
 >> Stay informed about: 5 in a row Tic Tac Toe 

hugo.alberto.DeleteThis@gmail.com wrote in news:1114091823.846692.153820
@g14g2000cwa.googlegroups.com:

> Thanks for the reply, i've search already the work of Victor Allis,
> but all i get is broken links. In my case i'm not really interested in
> a different search, the main objective of my work is to develop good
> heuristcs to the minimax wtih alpha beta pruning, and heuristics to
> prune irrelevant states of the board, eg. the states in the corners are
> irrevelant and therefore i will not generate them, this way i can
> reduce the space and time complexity, i'm trying to discuss things
> like this.

Part of Victor's work dealt with the kind of pruning you're talking
about, but I guess that doesn't do you any good if you can't access the
papers.

As far as developing good scoring heuristics:

If you happen to have lots (millions) of high-quality, scored board
positions available, look into M. Buro's "Sophisticated Features..."
paper. Look for it on citeseer. Also check out some of the papers that
reference his paper.

The disadvantage to automatic feature generation, besides needing a
training set, is that the algorithms are very poor at generalizing
(inducing) features; that is, they're extremely literal-minded. They
don't recognize that patterns of pieces can be shifted around the board,
rotated, mirrored, etc. As far as the algorithm is concerned, every
pattern is separate and unconnected to any other pattern. So what to
humans seems like one conceptual feature (like, say, "X threatens a win
next move") may take thousands or even millions of literal features for
the computer to express. Chances are, unless the training set is really
good, it's going to miss a lot of those literal expressions, leading to
inexplicable weaknesses in the resultant AI.

Without a training set, the state of the art is still just plain old
human expert knowledge codified into a complex set of board features.
Algorithms exist to automatically tune the weights of those features, but
they won't do you much good without an excellent feature set up front.
 >> Stay informed about: 5 in a row Tic Tac Toe 

i´ve just finish read the Victor Allis's thesis, it's great, it was
just what i was looking for, all that concept of threat-space search
will be of great help.
 >> Stay informed about: 5 in a row Tic Tac Toe