maximinLHD: Generate a maximin Latin-hypercube design (LHD)

Description

This function generates a LHD with large maximin distance.

Usage

maximinLHD(
  n,
  p,
  design = NULL,
  power = 2 * p,
  max.sa.iter = 1e+06,
  temp = 0,
  decay = 0.95,
  no.update.iter.max = 100,
  num.passes = 10,
  max.det.iter = 1e+06,
  method = "full",
  scaled = TRUE
)

Value

design: final design points.
total.iter: total number of swaps in the optimization.
criterion: final optimized criterion.
crit.hist: criterion history during the optimization process.

Arguments

n: design size.
p: design dimension.
design: an initial LHD. If design=NULL, a random LHD is generated.
power: the power used in the maximin objective function.
max.sa.iter: maximum number of swapping involved in the simulated annealing (SA) algorithm.
temp: initial temperature of the simulated annealing algorithm. If temp=0, it will be automatically determined.
decay: the temperature decay rate of simulated annealing.
no.update.iter.max: the maximum number of iterations where there is no update to the global optimum before SA stops.
num.passes: the maximum number of passes of the whole design matrix if deterministic swapping is used.
max.det.iter: maximum number of swapping involved in the deterministic swapping algorithm.
method: choice of "deterministic", "sa", or "full". If the method="full", the design is first optimized by SA and then deterministic swapping.
scaled: whether the design is scaled to unit hypercube. If scaled=FALSE, the design is represented by integer numbers from 1 to design size. Leave it as TRUE when no initial design is provided.

Details

maximinLHD generates a LHD or optimize an existing LHD to achieve large maximin distance by optimizing the reciprocal distance (see maximin.crit). The optimization details can be found in customLHD.

Examples

Run this code

# We show three different ways to use this function.
n = 20
p = 3
D.random = randomLHD(n, p)
# optimize over a random LHD by SA
D = maximinLHD(n, p, D.random, method='sa')
# optimize over a random LHD by deterministic swapping
D = maximinLHD(n, p, D.random, method='deterministic')
# Directly generate an optimized LHD for maximin criterion which goes
# through the above two optimization stages.
D = maximinLHD(n, p)

Run the code above in your browser using DataLab