Construct the quantitative portion of an initial design — magp_quantitative

Uses simulated annealing to improve a Latin hypercube under a maximin-style Euclidean-distance criterion. A candidate is created by swapping two values within one column, which preserves the Latin hypercube property.

Usage

magp_quantitative_design(
  n,
  q,
  p = 15L,
  maxit = 10000L,
  temp = 0.01,
  tmax = 10L,
  initial = NULL,
  seed = NULL
)

Arguments

n: Number of design runs. Must be at least two.
q: Number of quantitative variables. Must be positive.
p: Positive whole-number exponent controlling emphasis on the shortest pairwise distances.
maxit: Positive whole number of simulated-annealing iterations.
temp: Positive initial temperature passed to stats::optim().
tmax: Positive whole number of evaluations at each temperature.
initial: Optional n by q Latin hypercube with values in [0, 1].
seed: Optional nonnegative whole-number seed. When supplied, the function restores the caller's random-number state before returning.

Value

An object of class magp_quantitative_design, containing the optimized Latin hypercube, the starting design, criterion values, minimum distances, and search settings.

References

Kirkpatrick, S., Gelatt, C. D., and Vecchi, M. P. (1983). Optimization by Simulated Annealing. Science, 220, 671-680. doi:10.1126/science.220.4598.671 .

Examples

design <- magp_quantitative_design(
  n = 8,
  q = 4,
  maxit = 500,
  seed = 1
)
design$quantity
#>      quantity_1 quantity_2 quantity_3 quantity_4
#> [1,]     0.1875     0.8125     0.5625     0.9375
#> [2,]     0.9375     0.4375     0.0625     0.6875
#> [3,]     0.5625     0.9375     0.3125     0.4375
#> [4,]     0.4375     0.1875     0.4375     0.8125
#> [5,]     0.6875     0.6875     0.9375     0.5625
#> [6,]     0.3125     0.3125     0.1875     0.0625
#> [7,]     0.8125     0.0625     0.6875     0.1875
#> [8,]     0.0625     0.5625     0.8125     0.3125
design$criterion
#> [1] 1.659959