The minfx project is a Python package for numerical optimization. It provides a large collection of standard minimization algorithms, including the line search methods (steepest descent, back-and-forth coordinate descent, quasi-Newton BFGS, Newton, and Newton-CG), the trust-region methods (Cauchy point, dogleg, CG-Steihaug, and exact trust region), the conjugate gradient methods (Fletcher-Reeves, Polak-Ribiere, Polak-Ribiere +, and Hestenes-Stiefel), the miscellaneous methods (Grid search, Simplex, and Levenberg-Marquardt), and the augmented function constraint algorithms (logarithmic barrier and method of multipliers).
|Tags||Software Development Libraries Scientific/Engineering Python package|
|Operating Systems||OS Independent|
Release Notes: This is a minor feature release with improved documentation and support for sparseness in the grid search algorithm.
Release Notes: This release added Python 3 support and the logarithmic barrier augmented function constraint algorithm. All of the package, module, class, function, and method docstrings have been updated to Epydoc format, improving the online documentation. A few bugs have been eliminated, and the printouts have been regularised.
Release Notes: This release introduces a preliminary simulated annealing package based on scipy, and heavily modifies and improves the grid search algorithm.
Release Notes: Support for Python 2.6 has been added and a rare bug in the backtracking step selection subalgorithm has been removed.
Release Notes: This release involves the inevitable switch from Numeric python to numpy, a few improvements in how missing gradients and models with no parameters are handled, and a switch from GPLv2 to GPLv3.