PowellMethods is a Julia interface to some of the
derivative-free optimization algorithms by M.J.D. Powell:
-
COBYLA(Powell, 1994) accounts for general inequality constraints. -
NEWUOA(Powell, 2006) is for unconstrained optimization. -
BOBYQA(Powell, 2009) accounts for bound constraints on the variables.
This package is a wrapper over the implementations of these algorithms in the C
language in OptimPack libraries. Precompiled
versions of these libraries are provided by
OptimPack_jll
package.
For a more mature interface to M.J.D. Powell algorithms, I strongly recommend to use the PRIMA package.
The easiest way to install PowellMethods is via Julia registry
EmmtRegistry:
using Pkg
pkg"registry add General" # if not yet any registries
pkg"registry add https://github.com/emmt/EmmtRegistry"
pkg"add PowellMethods"-
M.J.D. Powell, "A direct search optimization method that models the objective and constraint functions by linear interpolation" in Advances in Optimization and Numerical Analysis Mathematics and Its Applications, vol. 275 (eds. Susana Gomez and Jean-Pierre Hennart), Kluwer Academic Publishers, pp. 51-67 (1994).
-
M.J.D. Powell, "The NEWUOA software for unconstrained minimization without derivatives" in Large-Scale Nonlinear Optimization, editors G. Di Pillo and M. Roma, Springer, pp. 255-297 (2006).
-
M.J.D. Powell, "The BOBYQA Algorithm for Bound Constrained Optimization Without Derivatives", Technical report, Department of Applied Mathematics and Theoretical Physics, University of Cambridge (2009).