Minfx 1.0.0.

Item posted by Edward d Auvergne <bugman> on Sat 08 Dec 2007 06:01:10 PM UTC.

Description

This is the initial release of minfx. The minfx project is a Python module library for numerical optimisation, being a large collection of minimisation algorithms. In the future it will have API interfaces for other programming languages added and will be ported to C for better numerical efficiency using the Lapack and Blas libraries.

This code originated as part of the relax project (http://nmr-relax.com or https://gna.org/projects/relax/) within the 'minimise/' directory. Minfx is complete, very stable, well tested, and has been spun off as its own project for the benefit of other scientific, analytical, or numerical projects.

Current optimisation algorithms

Line search methods: Steepest descent; Back-and-forth coordinate descent; Quasi-Newton BFGS; Newton; and Newton-CG.

Trust-region methods: Cauchy point; Dogleg; CG-Steihaug; and Exact trust region.

Conjugate gradient methods: Fletcher-Reeves; Polak-Ribiere; Polak-Ribiere +; and Hestenes-Stiefel.

Miscellaneous: Grid search; Simplex; and Levenberg-Marquardt.

The step selection subalgorithms include: Backtracking line search; Nocedal and Wright interpolation based line search; Nocedal and Wright line search for the Wolfe conditions; More and Thuente line search; and no line search.

The Hessian modification subalgorithms include: Unmodified Hessian; Eigenvalue modification; Cholesky with added multiple of the identity; the Gill, Murray, and Wright modified Cholesky algorithm; and the Schnabel and Eskow 1999 algorithm.

All methods can be constrained by the Method of Multipliers (also known as the Augmented Lagrangian).

Download

The software can be downloaded from http://download.gna.org/minfx/.