Solves the XOR. Is it enough?

[Kat007]

Hello,
I have a network that solves the XOR problem, however struggles with funcitons like z = y + x if I mix positive and negative values of x and y, or y=x^2 seems tough as well...
Is it enough to declare the network as fully working since it solves the XOR?
Plus any ideas why can't seem to get it to learn in 3D? It does ok if all inputs are positive, even if learning z = y^2 + x^2, but as soon as mix inputs with +ve and -ve again it strggles a lot.
Thank you

[kryptkat]

there is not enough information to get a good idea of what you are trying to do here.

graph it. that may or will help clear up what you are doing.

as for the 3d.... example if you have a circle that has 360 degrees and you give the variable a negative number.... a negative number is garbage from the point of view of the circle. negative number would be meaning to subtract from a starting point that is not specified here so it confuses the formula and has no idea what to do with the negative number. since you can only have coordinates given in positive numbers.

hope that cleared things up for you. meow.

[User Name:]

It depends on what you want it to do, whether it's enough or not. As to your problem with your NN not working, it could be that you need to add another layer(or more). My (limited) understanding leads me to believe that the layers must increase similarly to the complexity(nonlinear-ness) of the problem at hand.