modelSelector: Fast and exhaustive model selection in Python

The modelSelector Package offers a variety of functions for criterion based model selection in python.

Main Features

The Package supports:

• Various Model Selection Criteria:
  • AIC,
  • AICc,
  • BIC,
  • Mallow's Cp.
• Custom Models, as well as interfaces to linear and logistic regression models via scikit.learn.
• Standard selection algorithms, like
  • forward,
  • (efficient) backward,
  • exhaustive, and
  • efficient exhaustive.
• Visualization Routines for visualization of results.

Installation

From pip via

pip3 install modelSelector

or from GitHub via

git clone <insert repo url here>
pip3 install .

Getting started

Set up a linear regression model for given data (or alternatively use a wrapper around your own model) via

from modelSelector import LinearRegressionModel, AIC

# generate the full model, and fit it to data.
model = LinearRegressionModel('M_1111')
criterion = AIC()

# use any of our algorithms to perform the model selection 
result = backward_search(super_model=model,
                         criterion=criterion,
                         exclude_before_fitting=True,
                         data_x=data_x, 
                         data_y=data_y)

See the example notebook for further details and an overview over all supported algorithms.

Reference

Will be added after publication.