Skip to contents

Triangulates the cell centroids of one section (Delaunay), removes edges longer than max_edge, and counts for every cell how many of its remaining neighbours fall in each declared type group. This is the per-cell neighbourhood composition input of cellular-neighbourhood analyses.

Usage

NeighbourTypeCounts(
  cells,
  x = "x",
  y = "y",
  type = "type",
  levels,
  max_edge = Inf,
  engine = c("geos", "deldir"),
  duplicates = c("error", "isolate"),
  backend_contract = NULL,
  eps = 1e-09,
  index_base = 1L
)

Arguments

cells

Cell table of one section.

x, y, type

Column names of the coordinates and cell type.

levels

Named list of type groups (group name to cell types) or a character vector (each type its own group). Every cell type must belong to exactly one group. The order defines the count columns.

max_edge

Inclusive maximum edge length (same unit as x, y); Inf keeps all edges.

engine

"geos" or "deldir".

duplicates

"error" or "isolate".

backend_contract

Deprecated and ignored (see BuildExactDelaunayEdges()).

eps

Tolerance passed to the "deldir" engine.

index_base

Base of the integer type_index column (1 or 0).

Value

A data frame with one row per cell, in input order: type_group, type_index, x, y, degree and one integer count column per group. Attributes edges (kept undirected edges), full_edges (all Delaunay edges) and engine.

Details

Edge length is computed with the C library hypot() (via Mod()) on the double coordinates and compared inclusively (<= max_edge), as numpy.hypot does. In general position the Delaunay triangulation is unique, so every correct engine gives the same edges; with four or more co-circular points engines may choose different diagonals. Duplicate coordinates are refused by default; duplicates = "isolate" keeps the first occurrence as the graph vertex and gives later copies no neighbours (as Qhull's coplanar-point handling does).

Engines: "geos" (default) uses GEOS through sf (st_triangulate(..., bOnlyEdges = TRUE)); input coordinates are mapped back to cells exactly. "deldir" uses deldir::deldir() as the exact graph builder does; its run time grows roughly quadratically with the number of cells, so it is meant for cross-checks on small sections.

Examples

if (requireNamespace("sf", quietly = TRUE)) {
  set.seed(1)
  cells <- data.frame(x = runif(60, 0, 100), y = runif(60, 0, 100),
                      type = sample(c("a", "b", "c"), 60, replace = TRUE))
  out <- NeighbourTypeCounts(cells, levels = list(A = "a", BC = c("b", "c")),
                             max_edge = 20)
  head(out)
  attr(out, "edges")
}
#> [1] 125