ENH: Betareg rebased

[josef-pkt]

this is a rebased version of my branch https://github.com/josef-pkt/statsmodels/tree/betareg_2030_3 of PR #2030

I just did the rebase, I don't know what the status is

The unit tests still have some commented out yield statements without corresponding assert

    def test_methylation_precision(self):
        rslt = self.meth_fit
        #yield check_same, links.logit()(rslt.params[-2:]), expected_methylation_precision['Estimate'], 1e-3, "estimate"

When I use
assert_allclose(links.logit()(rslt.params[-2:]), expected_methylation_precision['Estimate'], 1e-3)
instead of the yield, then I get a large test failure

    def test_methylation_precision(self):
        rslt = self.meth_fit
>       assert_allclose(links.logit()(rslt.params[-2:]), expected_methylation_pr
ecision['Estimate'], 1e-3)
E       AssertionError:
E       Not equal to tolerance rtol=0.001, atol=0
E
E       (mismatch 100.0%)
E        x: array([ 36.043653, -36.043653])
E        y: array([ 8.22829, -0.03471])

So there are still problems with convergence when link_precision=links.identity().
(Which I don't think makes a lot of sense anyway, given the positivity constraint on variance/precision)
or there are problems in the unit test themselves.

[josef-pkt]

This pull request introduces 3 alerts - view on lgtm.com

new alerts:

• 3 for Potentially uninitialized local variable

---

Comment posted by lgtm.com

[coveralls]

Coverage decreased (-0.002%) to 82.169% when pulling 53a3518 on josef-pkt:betareg_rebased into c99558b on statsmodels:master.

[josef-pkt]

still failures, looks like it fails when scipy <= 0.16 and passes with newer scipy. (bfgs changes again?)

[josef-pkt]

Is this now generically implemented already in master?

see #2122
count_model uses result_class (which is missing an "s")