HGWR Kernel Interface — hgwr

Dispatch to one of the supported kernel functions and return the associated weights.

Usage

hgwr_kernel(d, bw, kernel = hgwr_kernel_names())

Arguments

d: Numeric vector or matrix of nonnegative distances.
bw: Positive bandwidth value, or a vector of bandwidths compatible with the columns of d.
kernel: Kernel name. Supported choices are "bisquare", "gaussian", "exponential", "tricube", and "boxcar".

Value

Numeric weights with the same shape as d.

Details

Kernel shape and bandwidth have different roles. Gaussian and exponential kernels have noncompact support; bisquare, tricube and boxcar are zero at and beyond the bandwidth. The exact kNN fitting engine supports only those three compact kernels. The Moran/LISA direct kernel choices form a smaller set; an explicit diagnostic weight matrix can be supplied separately. See vignette("lhwr-diagnostics-and-tuning", package = "LHWL") for all five formulas, overlaid weight curves and supported option combinations.

Examples

d <- c(0, 0.5, 1, 2)
hgwr_kernel(d, bw = 1, kernel = "bisquare")
hgwr_kernel(d, bw = 1, kernel = "gaussian")