sapphire

This repository contains companion code for the paper "SAPPHIRE: Preconditioned Stochastic Variance Reduction for Faster Large-Scale Statistical Learning".

Getting Started

1. Download the data:

   python download_data.py

2. Run experiments Options:

   • --run_all: Run all experiments
   • --exp_names: Specify experiment names (space-separated)

   Examples:

    python -m experiments.run_experiments --exp_names showcase
   
    python -m experiments.run_experiments --exp_names lasso_small logistic_small
   

   For a list of valid experiment names, see run_experiments.py (Line 78).

Results and Plotting

• Experiment results are saved as pickle files in the results folder.
• To generate figures, run the scripts in the plotting folder.
• Plots will be saved in results/plots.

Note: Pre-computed results and plots for all experiments are already included in the results folder.

Data Availability

All datasets (except for one) used for the experiments can be obtained running download_data.py. The exception is the UK Biobank dataset, which is not available for public distribution.