I am currently working on project to program a checkers program. My team has the two-player version almost completely done, but we are at a loss as to where to start concerning the neural net. We are programming in Java, and any advice given would be much appreciated.

First you should decide if neural networks are really what you want. It''s pretty easy to implement something like error back-propagation multilayer perceptron, but I think it wouldn''t be so easy to make complete AI using it.

And when you have programmed the algorithm you still have to train the net using maybe hundreds or thousands of examples. You will also have to take care avoiding local minimas etc.

Well, checkers AI still could be done with some kind of neural net. So if you really want to go for it I''d suggest you to buy or borrow some good book.

"Neural Networks : a Comprehensive Foundation" by Simon Haykin is pretty good, though it might involve a bit too much mathematics (Greens functions and stochastics for example) for some of you. I don''t know if there exist any game related neural network books.

-Ratsia

Hi,

I agree.

Many people think that it''s only to gather some datapatterns (in/output) and smash it into the NN and get a nice output. NN aren''t a magical box, it''s relatively easy to get a classification-net but when one wants to get a generalisation-net things become more complex.

Try to contemplate about what''s interesting for the net to know.
With a good dataset you''re closer to the goal.

I don''t know whether you''re in university, if you are, skip my first paragraph as you will most likely know that!!

Well, these are my opinions.


/Mankind gave birth to God.