Implement Sigmoid activation function in tensorflow and pytorch

[trevor-yap]

I would like to have to have sigmoid activation function as my variable is a probability.so I just want to bound my variable output between 0 and 1.
This will help increase the use case, instead of just use for regression.

[pobonomo]

Hi,
This is something that I have been playing with as Gurobi is improving it's capacity to solve nonlinear problems.
A model is not difficult to do but note that the size of the networks that we currently can solve is (in my limited tries) very small.

Do you have a case where it would be useful with a very small network?

[pobonomo]

The PR #473 would actually implement this. Will try to find some nice example to make it worth the merge.