fabian-sp / snspp

Stochastic Proximal Point algorithm for regularized statistical learning

SNSPP

SNSPP is a semismooth Newton stochastic proximal point method with variance reduction. The SNSPP method is implemented in snspp/solver/spp_solver and references therein.

Introduction

We aim for solving problems of the form

where the first part of the objective has the special form

This problem structure is common in statistical learning problems: each summand of f is the loss at one data sample and phi is a (convex), possibly nonsmooth regularizer. Note that for optimal performance f and phi should be Numba jitted classes.

Getting started

Install via

python setup.py

or in order to install in developer mode via

python setup.py clean --all develop clean --all

Requirements

The required packages are listed in requirements.txt. Here we list the versions of the most important packages that were used.

numpy==1.21.5
numba==0.55.1
sklearn==1.1.2
scipy==1.9.1
pandas==1.4.4
matplotlib==3.5.2
seaborn==0.11.2
csr==0.4.1