optimize_bw: Find positive minimizer with the input function

In gbasulto/kerdec: Kernel deconvolution methods

Description

Most methods used to find optimal bandwidths require to minimize a function (tipically MISE, AMISE, CV, ...). The argument ought to be positive. The fastest method that I empirically found was nlm. If the function is evaluated at a negative number or the output is not finite or not a number, the function will return the maximum possible number, so nlm can work better with the input function.

Usage

optimize_bw(f, h0, ...)

Arguments

f	Function to be minimized. Tipically a version of CV, MISE or AMISE.
h0	Initial bandwidth value.
...	Arguments passed to f besides the first argument, which must be the bandwidth.

Value

The output of the nlm function.

Author(s)

Guillermo Basulto-Elias

Examples

## Not run: 
## Bad initial point (it is negative).
kerdec:::optimize_bw(function(x) x^2, -5)

## End(Not run)

## Good initial point.
kerdec:::optimize_bw(function(x) x^2, 0.5)

gbasulto/kerdec documentation built on June 5, 2019, 10:58 a.m.