NLOP - Non-Linear OPtimization

Introduction

NLOP is an open-source C++ template library for non-linear optimization.

Motivation

I developed this project mainly because of my interest in convex optimization. I hope this will be useful for other convex optimization researchers for learning and teaching.

Features

• Thanks to the template programming, users can customize the data type to trade-off between speed and precision
• Autodiff function is provided, users don't have to give Jacobian formula manually
• Line search methods can be combined with step search method flexibly
• Users can enable log to have the optimization process information saved in an txt file, so that they can be used for visualization (a python script example is provided)

Usage

mkdir build
cd build
cmake ..
make -j4
./RosenbrockExample

Optimization method list

Linesearch methods

• Hooke & Jeeves's method
• steepest descent method
• conjuate gradient method
• Momentum
• Nesterov Momentum
• Adagrad
• RMSPro
• AdaDelta
• Adam

Newton's and quasi-Newton methods

• Newton's method
• DFP
• BFGS

Trust region methods

• Levenberg-Marquardt method

Stepsize searching method list

Accurate searching methods

• Golden section method
• Fibonacci method
• Dichotomous method

Inaccurate searching methods

• Armijo method
• Goldstein method
• Wolfe-Powell method

Requirements

• cmake
• Eigen3

Contacts

Please use Github issues to report bugs. If you have any question or advice, please feel free to contact Shuang Guo.