I was exploring fup´s CNeuralNet class, so that i would understand it, which i finally did... Then I decided to take on a "mission"...To make TicTacToe with semi-intelligent NN''s... I was playing around, making all those funcs which were needed, and guess what?? It worked! They(2 of ''em) actually play TTT! Perhaps not good, but they play! I´m som proud Anyway, would it improve the NN if i changed the number of layers to, say 10 and the number of neurons to say 81(now i have 1 layer with 9 neurons, 9 inputs and one output)... Or perhaps i should make 9 outputs, and the one with the highest value is the "box to check"...

Sure: http://www.gameai.com/buckland.html

I´m considering making the TTT complete and putting it on Open Source...If it´s OK with fup that i use his CNeuralNet...

In general, more neurons means the NN has less ability to generalize to similar situations and is more optimized to a particular training set and basically incompetent outside of that training set.

i did this:

#define INPUTS 9
#define OUTPUTS 1
#define HIDDEN_LAYERS 0
#define NEURONS_PER_LAYER 9

Which landed me at 90 neurons...

It appears that they became...faster...
And there was also a dramatic decrease in ties...
From 90% ties to about 10%, which pretty much says it all