Implementing the softmax activation function

[Kerollmops]

I saw that I will need the softmax activation function for my basic word2vec neural network. The thing is the softmax activation function needs the whole list of output values and the Activation trait does only know about the current output value. I would say that I must change the trait to give the whole list of values somehow but I find it not performant enough if I continue to call it on each output value, computing the exponential sum every-time.

[c0dearm]

Hey! Sorry for the late response.

I've just released a new version of the library which basically changes it completely so probably the issue is not relevant anymore.

Adding new operations/layers should be easy peasy now.