Build cost matrix for matching — build_cost

This is the main entry point for distance computation.

Usage

build_cost_matrix(
  left,
  right,
  vars,
  distance = "euclidean",
  weights = NULL,
  scale = FALSE,
  sigma = NULL,
  memory_mode = "auto",
  caller_supports_lazy = TRUE,
  caller_supports_implicit = caller_supports_lazy
)

Arguments

memory_mode: One of "auto" (default), "dense", "lazy" or "implicit". "auto" warns (or, when the caller supports it, switches) when the dense matrix would consume a large fraction of free system RAM. memory_mode = "lazy" returns a lazy_cost_spec instead of a matrix when the calling path and distance metric support it (built-in metrics via assignment() with method = "jv"/"auction"); otherwise it errors clearly rather than silently falling back to dense. memory_mode = "implicit" returns the same specification marked for the edge-generation loop, which solves it without building the pair set at all.
caller_supports_lazy: Whether the calling code path can actually consume a lazy_cost_spec result for a solve over every pair. Defaults to TRUE; a caller with no such solve (e.g. full_match(), whose flow over every pair would hold every pair) passes FALSE so memory_mode = "auto" never promotes to lazy for it, and an explicit memory_mode = "lazy" request errors clearly instead of returning a lazy_cost_spec the caller cannot use.
caller_supports_implicit: Whether the calling path's design is one the edge-generation loop solves. Defaults to whatever the caller says about lazy, since the loop reads the same specification.

Value

Numeric matrix of distances with optional scaling/weights applied.