models: hierarchical logistic regression

[rlouf]

I am writing documentation for the various models (a bit like scikit-learn's user guides), and I am struggling to understand HierarchicalLogisticRegression:

• The number of categories can be > 2, yet the probability of belonging to a category is given by a Bernouilli distribution;
• What is the cats variable useful for?

I would replace Bernoulli with Categorical and write directly temp = alpha + T.sum(beta*model_input, 1) or something like that. But maybe there is something I don't understand.

[parsing-science]

Hi, did you watch the PyData NYC talk I gave about this model? I think that should clear up some of these issues.