This is the accompanying code for our paper Grouped Variable Selection with Discrete Optimization: Computational and Statistical Perspectives.
This repo contains code for (i) approximate algorithms based on coordinate descent and combinatorial local search, and (ii) exact algorithms based on a custom branch-and-bound algorithm.
To get started please refer to Demo.ipynb
The package is written in Python 3. It requires the following prerequisites:
- numpy
- scipy
- numba
- gurobi (only needed for the BnB algorithm)
See the Jupyter notebook Demo.ipynb for a demonstration on how to use the different algorithms.