wfgEval: wfgEval and wfgWrap

In tudob/wfg: WFG toolkit implementation package

Description

wfgEval evaluates a test function (conforming to a specification) at a given point.
wfgWrap creates a functor (conforming to a specification) without evaluating it.
wfgEval can be directly executed e.g. in a loop trying various parameters.
wfgWrap returned functor can be given to optimization algorithms.

Usage

wfgEval(z, num.objectives, spec, z.maxes = NA, num.pos.rel = NA,
  degen = FALSE, scales = NA, dist = NA, noise = 0,
  true.front = FALSE)

wfgWrap(num.objectives, spec, z.maxes = NA, num.pos.rel = NA,
  degen = FALSE, scales = NA, dist = NA, noise = 0,
  true.front = FALSE)

Arguments

z	The point at which to evaluate the test function.
num.objectives	The size of the output vector. (|out|<=|in|)
spec	The specification of transformations, shapes and their parameters. specifically a vector or list with:
z.maxes	This and all following parameters are optional. (If you use any of them, that you place them before giving the spec.) A vector containing the maximum for each in-dimension (the minima are always 0). default is all 1.
num.pos.rel	The number of position-related parameters. For details refer to the paper (variable k).
degen	Whether to degenerate the pareto frontier. Normally it is a hyperplane, this degenerates it by 1 dimension.
scales	A vector of factors for the objectives.
dist	A single distance scaling factor. In principle moving the evaluations away from the true pareto frontier.
noise	The standard deviation of the multivariate normal distributed noise added on top.
true.front	Whether to instead return the point on the true pareto frontier corresponding to the input.

Value

wfgEval: The vector of objective values. wfgWrap: a functor that can be given to optimization algorithms.

Examples

v2 = c(0.1, 0.2) # example input vectors
v3 = c(0.1, 0.2, 0.3)
v4 = c(0.1, 0.2, 0.3, 0.4)
v5 = c(0.1, 0.2, 0.3, 0.4, 0.5)

wfgEval(v3, 3, c(sConvex))
       # Specification to only evaluate the convex shape, no transformations.

wfgEval(v3, 3, degen=TRUE, c(sConvex))
       # It is recommended placing any optional parameters in front of the spec (which can be long).

wfgEval(v3, 3, c(sLinear, 2, sMixed))
       # The first parameter is the number of entries to apply the front to. the default is to apply to all remaining entries, except for mixed and disconnected fronts which can only transform a single entry.

wfgEval(v3, 3, c(sDisc, NA, 1.3, sLinear))
       # Order-based parameters can be given, in this case the overall shape is 1.3, the other parameters of the disconnected shape remain their defaults.

wfgEval(v3, 3, c(tFlat, 3, 0.9, 0.2, 0.5, sLinear))
       # A transformation with order-based parameters (number of entries, value, from, to).

wfgEval(v3, 3, c(tDecept, aper=0.25, sLinear))
       # Named parameter (aperture size), all other parameters stay on default.

wfgEval(v3, 3, c(tPoly, alpha=0.02, tDecept, NA, 0.35, 0.001, 0.05, tMulti, 1, num.minima=30, tNonsep, sConvex, 1, sLinear))
       # Four transformations and two shapes. In this example all transformations are applied to all entries either by using named parameters or in positional parameters by skipping the apply-length with NA.

functor = wfgWrap(2, c(sLinear))
       # Compared to wfgEval the only difference in arguments is, that the first argument - the point z - is not given.

tudob/wfg documentation built on June 1, 2019, 2:54 a.m.