q.to.p: Real-to-multifrequency transformation

In mixstock: functions for mixed stock analysis

Description

Transform a vector of n real-valued variables in (-Inf,Inf) [or (0,1)] to a vector of n+1 variables in (0,1) that sum to 1, or vice versa.

Usage

q.to.p(q, transf="full")
p.to.q(p, transf="full")

Arguments

q	Unconstrained/transformed values: vector of n numeric values in (-Inf,Inf) [if transf="full"] or (0,1) [if transf="part"]
p	Vector of n+1 numeric values in (0,1) that sum to 1
transf	(character) "full": use arctan transform to transform (-Inf,Inf) to (0,1) or vice versa; "part": don't; "none"; no transform

Details

Essentially, this is a transformation from an unconstrained set of variables to a bounded, constrained set of variables. If contin is TRUE, an arctan transformation (v <-> atan(v)/pi+0.5) is used to transform (-Inf,Inf) to (0,1) or vice versa. In either case, the correlated set of variables (which sum to 1) is transformed to an unconstrained set by taking each variable to be a remainder: x[1]=x[1], x[2]=x[2]/(1-x[1]), and so forth.

Value

Vector of transformed values.

Note

This transformation is designed to deal with the problems of bounded optimization and constraints. It actually behaves quite badly because small values are transformed to large negative values, messing up the uniform scaling of the parameters. Now that the bounded optimization of optim has improved, contin="full" may not be a good idea. It's not clear whether the other transformation (remainders) is better or worse than just optimizing on the first (n-1) components and assuming that the last frequency equals one minus the sum of the rest.

Author(s)

Ben Bolker

Examples

  p.to.q(c(0.3,0.3,0.4))
  p.to.q(c(0.3,0.3,0.4),transf="part")
  q.to.p(c(-4,3))
  q.to.p(c(0,0))
  q.to.p(c(0.5,0.5),transf="part")

mixstock documentation built on May 2, 2019, 6:48 p.m.