Generate multiple mutually-correlated variables at a target correlation level

Builds n_inputs variables, each nLUT values long, all pairwise-correlated at (approximately) rho, then rescales each to its own [MinRange, MaxRange] bound. Useful for LUTs where several traits should co-vary (e.g. LAI and canopy height) rather than being drawn independently.

Usage

getCor(
  n_inputs = NULL,
  nLUT = 100,
  distribution = "Uniform",
  setseed = 123,
  rho = NULL,
  Varnames = NULL,
  MinRange = NULL,
  MaxRange = NULL
)

Arguments

n_inputs

integer. How many variables to generate.

nLUT

integer. How many values per variable (LUT rows). Default 100.

distribution

character. "Uniform" or "Normal". Default "Uniform".

setseed

integer. Random seed. Default 123.

rho

numeric (-1 to 1). Target pairwise correlation between every pair of variables.

Varnames

character vector, length n_inputs. Column names for the output; if NULL, columns are named Var_1, Var_2, ...

MinRange

numeric vector, length n_inputs. Minimum of each variable's output range.

MaxRange

numeric vector, length n_inputs. Maximum of each variable's output range.

Value

A list: LUT (data.frame, nLUT rows x n_inputs columns) and Covarianza (the realized correlation matrix of LUT, for checking how close the sample came to the requested rho).

Examples

out <- getCor(n_inputs = 2, nLUT = 200, distribution = "Uniform", rho = 0.7,
               Varnames = c("LAI", "Height"), MinRange = c(0.5, 2), MaxRange = c(7, 30))
#> Generating a Uniform distribution for all correlated inputs ...
cor(out$LUT$LAI, out$LUT$Height)  # close to 0.7
#> [1] 0.7273228