Tic Tac Toe Neural Network

[glo]

Hi.

I want to implement a neural network into this simple Tic Tac Toe game I wrote and basically teach it to play decently. I mainly want to understand neural networks and Tic Tac Toe seems like the right entry level.

I take it the inputs should be the 9 board squares, and output should be the board square number to mark next.
If the round is won, all choices made throughout the round should have their weights strengthened, but where is all this stored? Do I create a table that maps every possible board possibility and its weight values for every possible move you can perform at that position?

Will appreciate clarifying this.
Thanks,
glo

[guesst]

Is there a class somewhere that's assigning tic (http://cboard.cprogramming.com/showthread.php?t=103715&highlight=tic+tac+toe)-tac (http://cboard.cprogramming.com/showthread.php?t=103861&highlight=tic+tac+toe)-toe (http://cboard.cprogramming.com/showthread.php?t=101567&highlight=tic+tac+toe) to it's students and they're all coming here??

[glo]

Again, I'm mainly trying to understand neural network. Choosing Tic Tac Toe was rather random, but you would have to admit it's a pretty basic game to experiment on.
In any case, feel free to advise more generically about neural networks and how can I implement them.

[Magos]

Perhaps better to use 9 in and 9 out (each representing a percentage of good choice to pick the cell as next move, the cell with highest value should be picked). Neural networks doesn't deal well with integral values as far as I remember.
Didn't read much but this site seems to explain some parts well:
http://www.adit.co.uk/html/neural_networks.html

[abachler]

If your goal is to make a tic tac toe game then neural networks may not be the best choice. Tic tac toe is an example of a 'solved problem'. The number of attainable positions is finite, known, and each choice has a definite best move although several moves may be equally good. It is useful for learning statistical learning methods.

If your goal is to learn neural networks then tic tac toe may not be the best choice. Typically XOR is used as the 'hello world' problem. Again this is because it is a known problem and every case can be tested and has a known solution.


I'm not saying tic tac toe cant be done with neural networks, just that you should probably learn to roll over on your stomach before you learn to crawl.