ENH: Segmented knots rebased

[josef-pkt]

rebased version of #2677

removed last commit which was compat fixes for numpy < 1.9 (in percentile)

status not clear see old comments https://github.com/statsmodels/statsmodels/pull/2677#issuecomment-226978367 predict and results info incomplete or missing

old notebook https://github.com/josef-pkt/misc/blob/segmented/notebooks/ex_segmented_regression_sm.ipynb

segmented is in base get_objective needs llf as alternative to ssr then we could try a poisson or similar example

[pep8speaks]

Hello @josef-pkt! Thanks for opening this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file statsmodels/base/_segmented.py:

Line 177:9: E306 expected 1 blank line before a nested definition, found 0

[josef-pkt]

does this use (effective) degrees of freedom corresponding to the number of segments?

[lgtm-com[bot]]

This pull request introduces 1 alert when merging a0b57b41417c8eb93c0724375e11fad344a4f0ce into 152e27dcc667a423f798810a1b0fee2f6340f73d - view on LGTM.com

new alerts:

• 1 for Variable defined multiple times

[josef-pkt]

unit tests pass, pep-8 style failures

[josef-pkt]

not clear from a quick look at this PR:

usage for other models than OLS? The module is base/_segmented.py and seems to be designed for generically for models. However, I don't see a non-OLS example here. For OLS only, it would fit better in regression.special_model(s)