Feature request: multinomial logistic regression

[DilumAluthge]

Are there any plans to implement multinomial logistic regression?

Specifically, I am referring to the situation of a categorical output variable that has strictly greater than two possible classes. The classes are not ordered.

[DilumAluthge]

It looks like @Nosferican has implemented this in Econometrics.jl. Would it be possible to merge some of that work into GLM.jl?

[Nosferican]

It would require to change a few of the internals though. The way I fit it in Econometrics is through a vector generalized linear model. It would require the internals to allow for vector responses rather than a single response vector. I believe that is the biggest change. In terms of outputs, the parameters are per response so it would have response_predictor.

[jbrea]

I'd also be interested in this feature (and JuliaStats/Lasso.jl#8).

[andreasnoack]

Dup of #206