Stochastic Variance-Reduced Iterative Hard Thresholding in Graph Sparsity Optimization

Overview

This repository houses the code for "Qianqian Tong, Derek Fox, and Samuel Hernandez, Stochastic Variance-Reduced Iterative Hard Thresholding in Graph Sparsity Optimization".

This code is based off of the excellent work found in: "Baojian Zhou, Feng Chen, and Yiming Ying, Stochastic Iterative Hard Thresholding for Graph-structured Sparsity Optimization, ICML, 2019".

Our code is written in Python. The implementation of head and tail projection are from the aforementioned paper, written in C11.

Instructions

Below are the installation instructions:

0.  Clone the code to your machine: git clone https://github.com/Derek-Fox/graph-scsg-iht.git

1.  Install python2.7 and gcc. I highly suggest setting up a conda/venv environment to make this easy!!

2.  Install numpy, scikit-learn, matplotlib, and networkx.

3.  After the above three steps, run: python setup.py build_ext --inplace.

After step 3, it will generate a sparse_module.so file, which will allow for the head and tail projections from Zhou et al. to run properly.

Run Experiments

To run tune_params.py, simply run python tune_params.py in your terminal. In the main method, you can comment out the function calls to determine which parameters to tune/generate figures for.

To run exp_sr_test02, call python exp_sr_test02.py run_test <num_cpus> with the number of cpus to use.

To run exp_bc_run, call python exp_bc_run.py <num_cpus> <trial_idx_start> <trial_idx_end> with the number of cpus, and the beginning and end trial numbers to run. For instance, to run 20 trials, you would call python exp_bc_run.py <num_cpus> 0 20 with the appropriate number of cpus for your system.