Sigmoid or Gaussian output

Hi, like you, I agree that an RBF classifier is given less credit than it deserves. 

I've been trying to read your code without any knowledge if Theano. (I'm a Keras newbie).

It appears you have used the sigmoid similarity function instead of the gaussian similarity. 

I'm trying to build a very simple three-layer RBF network.

I'd would be delighted to hear back from you.
