Skip to contents

select_resolution() reads one criterion at a time. This puts the count criteria in one table: the level each prefers, the flat region around it, and whether a ladder bound is doing the choosing rather than the criterion. The closing line gives the levels that lie in every flat region, the cell counts no criterion objects to.

Usage

# S3 method for class 'resolution_profile'
summary(object, criteria = NULL, tol = 0.02, ...)

Arguments

object

A resolution_profile().

criteria

Character vector, any of "cp", "reliability", "elbow", "moran_z", "loo". Default: every one of "cp", "reliability", "elbow" and "loo" that is finite at some level (loo is a column of a profile built with estimate; a column of that name on a profile built without one is read only when named here). "moran_z", a diagnostic, is read only when named here. "size" and "auto" are not criteria of the table: a size sets the count and has no flat region to compare.

tol

Passed to select_resolution() for the flat region. Default 0.02.

...

Ignored.

Value

A data.frame of class resolution_summary, one row per criterion in the order given, with columns criterion, best, flat_min, flat_max, n_flat, value (the optimum itself, on that criterion's own scale and so not comparable across rows, which is why the print method leaves it out), at_floor, at_ceiling, edge (which bound an edge optimum sits on, as select_resolution() reports it; NA when interior) and structured (as select_resolution() reports it: FALSE for a cp that one cell matches, and where the cells beat one cell but the F test does not tell their means from one mean or, for cp, the profile's variogram found no spatial structure). Attributes: bands (a named list holding each criterion's flat region in full, since flat_min and flat_max are only its ends and the region can have holes in it), common (the levels in every flat region of the count criteria, moran_z left out; an integer vector that is empty when the regions do not overlap), scored (the levels each criterion returned a finite value at, which is what decides whether a gap in a flat region is a rejection or a level nothing was computed at), ladder (every level on the profile), n_levels, tol and variable (what the criteria were scored on), and, only when some row has structured FALSE although its cells beat one cell by more than tol, not_identified (a character vector named by criterion that says why: the F test, or for cp the profile's variogram). The print method recomputes the closing comparison from bands, so a row subset of the result prints its own comparison.

Details

Read the table with what each criterion returns in mind (see resolution_profile()): reliability is read at or above the range floor and often returns it, elbow is absent unless the points cluster, and cp's optimum is often the support ceiling. moran_z is a diagnostic of the cell means and not a count rule, so it is not in the default set; named in criteria it gets a row, and it is left out of the closing comparison.

The criterion curves are not monotone, so a flat region can skip rungs of the ladder, and the table lists the levels it accepts ("26, 31") rather than a range ("26 to 31") that would include the levels it rejected.

The levels common to every flat region are often none. The criteria answer different questions: how well the cells represent the field (\(C_p\)), whether the cell values are distinguishable from noise (reliability), and where the within-cluster sum of squares bends (elbow). An empty set means the criteria favour different counts, and the print gives the spread between the picks.

A flat region can be wide, and the choice within it is the analyst's; plot.resolution_profile() draws the curves the regions were read from.

A profile whose ladder is one size (obs_per_cell set to one number, or the default profile) has one level, and the size set it. The table then has one row per criterion that is finite at that level, and the line under it says that the count is set by the size and that no criterion chooses among levels. The result carries the size as attr(, "obs_per_cell").

See also

select_resolution() for one criterion, with the per-level values attached; plot.resolution_profile() for the curves behind these flat regions.

Other resolution: determine_optimal_levels(), plot.resolution_profile(), resolution_profile(), select_resolution()

Examples

library(sf)
# An exponential field with range parameter 200 (true effective range
# 600 m) on a 1 km square, with a nugget of 0.6 on a unit sill.
set.seed(4)
n <- 400
xy <- data.frame(x = 5e5 + runif(n, 0, 1000), y = 5e6 + runif(n, 0, 1000))
D  <- as.matrix(dist(xy))
xy$z <- as.numeric(t(chol(exp(-D / 200) + diag(0.6, n))) %*% rnorm(n))
pts <- st_as_sf(xy, coords = c("x", "y"), crs = 32632)
prof <- resolution_profile(pts, response_var = "z", n_levels = 12,
                           estimate = NULL)

s <- summary(prof)
s
#> Resolution picks: 2 criteria over 12 levels (2 to 44 cells)
#> 
#>   cells: density-weighted k-means++ starts (power 1, 8 neighbours)
#> 
#>    criterion best  flat region levels in flat region
#>           cp   19 19 to 25, 44                     3
#>  reliability    4       4 to 6                     3
#> 
#>   reliability: the optimum is the range floor (area / range^2).
#>   There the bound is choosing, not the criterion.
#> 
#>   picks span 4 to 19 cells (4.8x)
#>   no level is in every flat region: the criteria disagree over the
#>   whole ladder. plot() draws the curves they were read from.
attr(s, "common")   # the levels every flat region accepts, if any
#> integer(0)
attr(s, "bands")    # each criterion's flat region in full
#> $cp
#> [1] 19 25 44
#> 
#> $reliability
#> [1] 4 5 6
#>