Yutong-Dai / S-PStorm

Source code for the work A Variance-Reduced and Stabilized Proximal Stochastic Gradient Method with Support Identification Guarantees for Structured Optimization accepted by AISTATS2023.

S-PStorm

Source code for the work A Variance-Reduced and Stabilized Proximal Stochastic Gradient Method with Support Identification Guarantees for Structured Optimization accepted by AISTATS2023.

This repo contains implementations of a collection of stochastic first-order methods, including ProxSVRG, SAGA, RDA, PStorm, and S-PStorm. The repo contains two main directories, src and test. src/solvers contain the source code for all algorithm implementations. test directory contains the scripts necessary to reproduce the results reported in the paper.

Usgae of the code

Data Preparation

Navigate to the test/data_prep directory and do the following steps.

1. Run bash download.sh to download 10 datasets.
2. Run bash process.sh to perform data preprocessing.
3. Run python compute_Lip to get the estimate of the Lipschitz constants.

Perform tests

1. Navigate to the directory: cd test/bash.
2. Generate the bash scripts: python create_bash.py
3. Run the command bash submit and experiments will run in the background. The logs on each run can be found at test/bash/log and the results will be saved at test/experiments