IGA: An Interactive Greedy Approach to Sparsity Learning in High Dimensions

Introduction

Interactive Greedy Algorithm (IGA) is a forward-backward stepwise variable/group selection algorithm for sparsity learning in high dimensional regression problems. The description of IGA algorithm and its applications are given in the papers shown below.

This repo contains a MATLAB demo and implementation of IGA algorithm for sparse linear regression and logistic regression.

License

IGA is released under the GPL-3 License (refer to the LICENSE file for details).

Papers regarding IGA

@article{qian2022adapative,
title={Adaptive algorithm for multi-armed bandit problem with high-dimensional covariates},
author={Qian, Wei and Ing, Ching-Kang and Liu, Ji},
journal={Journal of the American Statistical Association, in press},
year={2022+}
}

@article{qian2019interactive,
title={An interactive greedy approach to group sparsity in high dimensions},
author={Qian, Wei and Li, Wending and Sogawa, Yasuhiro and Fujimaki, Ryohei and Yang, Xitong and Liu, Ji},
journal={Technometrics},
year={2019},
volume={61},
issue={3}
}

Running Demo:

0. Add the folder ./MATLAB_fcn to MATLAB search path
1. Use demo1.m to run a high-dimensional linear regression example with IGA.
2. Use demo2.m to run a high-dimensional logistic regression example with IGA.