Stepwise regression fits a logistic regression model in which the choice of predictive variables is carried out by an automatic forward stepwise procedure. The stepwise selection process can be visualized and shows the saturation in terms of AUC. Furthermore it is possible to use early stopping to get an parsimonious model.

The package can be installed from GitHub using pip:

pip install https://github.com/Mathias-Kreis/stepwise-regression/stepwise-regression.zip

If you have issues to install the package, it is possible to just download stepwise-regression.zip and navigate via anaconda prompt to the unpacked package location and install it with

pip install .

To use the functionality of stepwise-regression you just have to import the main function Logitstep as follows:

from stepreg.stepreg import Logitstep

Here you can see the Logitstep function with input parameter:

logit_results, logit_model, steplist = Logitstep(
                                                  y,
                                                  X,
                                                  maxsteps=30,
                                                  p=0.05,
                                                  force=["intercept"],
                                                  plot=True,
                                                  early_stopping=False,
                                                  patience=10,
                                                  tol=0.001,
                                                  verbosity=1
                                                  )

The function returns three Objects: the model summary, the estimated model itself and a list of selected variables. Having the function ready for action, just give it a try, but bear in mind that you prepare your data properly.

You can find an example in the main branch which models Fraud data that can be find on kaggle. In particular you can find a possible data prepartion before running Logitstep.

The package is still under construction. I am thankful for hints and comments!