Rock Paper Scissors

I am working on an assignment to play a game of Rock Paper Scissors
against a human opponent. I am using a C4.5 decision tree induction to
evaluate past moves and would like some help.

1) Does anyone have experience with WEKA? Building a CSV and calling
the C4.5 commandline after each move is horribly inefficient and a pain
in the butt.

2) Should I initially train the tree using random moves for several
iterations and then start "playing" using the tree?

3) What features should I use? I am operating on the assumption that
humans do not wisely keep track of their opponents decision making
process beyond the past few moves. Given this, I think the moves my bot
has made are generally arbitrary and the features should simply be
based on the last several moves. Does this render my tree into a simple
statistical analysis?

Any help or insight is greatly appreciated. I've been getting about
50/50 after several dozen iterations.

Chris

chris.RemoveThis@siokaos.org wrote:
> 3) What features should I use? I am operating on the assumption that
> humans do not wisely keep track of their opponents decision making
> process beyond the past few moves. Given this, I think the moves my bot
> has made are generally arbitrary and the features should simply be
> based on the last several moves. Does this render my tree into a simple
> statistical analysis?

yep, it looks pretty much like this. whatever model of human reaction
you implement, you will end up with probabilities of his moves; you
then should assume the move with highest prob, and react accordingly.