Stepshift is a package that implements the stepshifting algorithm described in appendix A of
Hegre et al. (2020).
Stepshift is currently only distributed as a source distribution, which means that the end user needs a C compiler. This means that OSX users need to have Xcode installed on their system before proceeding. In addition, the numpy requirement is quite strict, since stepshift uses the Numpy C API via Cython.
Install by running:
pip install stepshift
Stepshift has a module called stepshift.views which contains a class called
StepshiftedModels. This class wraps the stepshifting procedure, exposing a
simple, Scikit-Learn-like (but not equivalent) API. The model takes three
arguments: A scikit learn estimator, a list containing integers, which denotes
the steps, and a string variable which is the name of the dependent variable:
from sklearn.linear_model import LogisticRegression
from stepshift.views import StepshiftedModels
mdl = StepshiftedModels(LogisticRegression(),[1,2,3,4,5,6,7,8],"outcome")