Open
Description
We already have binomial_logit_glm
and categorical_logit_glm
Multinomial distribution generalizes both Binomial and Categorical, is still in the exponential family and the logit (softmax) parametrization seem natural - we already have multinomial_logit
. So multinomial_logit_glm
is a missing piece to be completed.
I'm actually using this in an application (modelling counts of sold products each day), where I simply iterate multinomial_logit
, but it would be cool to have an optimized single-call function.
Related issue: #1964
Metadata
Metadata
Assignees
Labels
No labels