One row per unit of work and element covered, for any sequence from
solve_by_sample(), solve_by_region(), solve_recursive() or
solve_guided().
Usage
sequence_coverage(x, ...)
# Default S3 method
sequence_coverage(x, ...)
# S3 method for class 'by_sample'
sequence_coverage(x, ...)
# S3 method for class 'by_region'
sequence_coverage(x, ...)
# S3 method for class 'recursive'
sequence_coverage(x, ...)
# S3 method for class 'guided'
sequence_coverage(x, ...)Details
The dim column names the scale the units are drawn on — "timeslice" for
a calendar sample, "region" for a spatial one, "year" for a horizon
walk — so a caller can join the result onto the matching
timescales::Calendar or geoscales::GeoScale.
role separates what a unit decided from what it merely saw: a
recursive step's look-ahead milestones are solved for guidance and their
decisions discarded, so they are coverage of a different kind and are
labelled "lookahead". Everything else is "covered".
count is the multiplicity of an element within its unit. It is 1 except
for a bootstrap calendar sample, where a timeslice drawn twice carries a
doubled share — overlap within a unit rather than between units.
Reading the result
Two questions are worth asking of it directly. Whether the units partition
the scale: any(duplicated(res$key[res$role == "covered"])) is TRUE when
two units claim the same element. And what was missed: compare res$key
against the scale's own members — the elements absent from the frame were
never solved, which is what a gap in an ensemble looks like.
Examples
if (FALSE) { # \dontrun{
cv <- sequence_coverage(bs)
# did the samples tile the calendar, or overlap?
any(duplicated(cv$key[cv$role == "covered"]))
} # }