Regularization of \beta_0

[RogerioOMDS]

ln 37 - 43, on sthocastic_gradient.jl.

In the current model, the algorithm penalizes the \beta_0 argument on regularization, what we don't have to do in a theorecally point of view. So, it's necessary to separate internally this value to don't suffer this penalization with all the other values.