I don't know of any.
There are recipes floating around on stackoverflow, and IIRC in our
issues, but no ready made functions in a package, AFAIK.
Stepwise selection doesn't have a good reputation, the properties of
it are not very good.
A better alternative is to use penalized estimation that includes
variable selection like LASSO.
The two related version that I worked on are ultrahigh screening and
all subset regression.
Neither, are in the public part of statsmodels.
Ultrahigh screening is when we have potentially many more variables
than observations, and combines forward selection of several variables
at the same time, with variable dropping based on SCAD penalization.
It has been in statsmodels for some time but still without official
model,
https://github.com/statsmodels/statsmodels/blob/master/statsmodels/base/tests/test_screening.py
If we have only a small number of explanatory variables, then we can
compute regression for all subsets of variables and choose with aic,
bic or similar.
Efficient code for that is still waiting in a PR.
Josef
> --
> You received this message because you are subscribed to the Google Groups "pystatsmodels" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to
pystatsmodel...@googlegroups.com.
> To view this discussion on the web visit
https://groups.google.com/d/msgid/pystatsmodels/31f135a0-f489-495c-a1c6-8c6494d14aadn%40googlegroups.com.