Crate newton_sos

Crate newton_sos 

Source
Expand description

The newton_sos crate defines and solves optimization problems of the form: $$\max_{c\in\mathbb{R}, B \in \mathbb{S}^n_+} c - \lambda \text{Tr}(B) + t \log \det (B) \qquad \text{s.t. }\quad f_i - c = \Phi_i^T B \Phi_i, \:\:\forall i\in[\![1, N]\!]$$ using a damped Newton method. Such problems arise from sum-of-squares optimization, especially in the Kernel Sum-of-Squares (KernelSOS) framework.

§Overview

The main components of the crate are:

  • problem::Problem: A struct representing the optimization problem, including data and parameters.
  • solver::solve: A function to solve the optimization problem.

§Feature Flags

  • python: Enables Python bindings using PyO3.

Modules§

problem
Defines the optimization problem structure, as well as methods for computing the features matrix and kernel matrix.
solver
Defines the main solver algorithms for the optimization problem.