BUG: MNLogit loglikeobs, jac

[josef-pkt]

in loglike, d and logprob, are 2d. I think to get loglikeobs we should add over axis=1

return (d * logprob).sum(1)  

I guess jac has a similar mistake.

The test uses sum() and so doesn't realize that loglikeobs has the wrong shape.

[josef-pkt]

I thought I fixed this but cannot find where (or if) right now.

[josef-pkt]

I don't find a bugfix, but this looks all correct

was added in
7302d53 #147

>>> np.max(np.abs(mlogit_res.model.jac(mlogit_res.params).sum(0) - mlogit_res.model.score(mlogit_res.params)))
1.0844797282416607e-13

>>> np.max(np.abs(mlogit_res.model.loglikeobs(mlogit_res.params).sum() - mlogit_res.model.loglike(mlogit_res.params)))
0.0