Keep only the leaf timeslices whose timeframe label is in labels
— the time-side mirror of geoscales::filter_geoscale(). Shares are
NOT renormalized: the result is a partial-year calendar whose
meta$year_fraction is set to the surviving sum(share) (use
calendar_share() when normalized shares are needed). Level
vocabularies are subset to the surviving labels.
Usage
filter_calendar(x, timeframe, labels)
# S3 method for class 'Calendar'
x[i, j, ...]
# S3 method for class '`timescales::Calendar`'
x[i, j, ...]Arguments
- x
A Calendar (the
[method's object).- timeframe
Timeframe the labels live at.
- labels
Labels at
timeframeto keep.- i, j
x[i, j]isfilter_calendar(x, timeframe = i, labels = j).- ...
Ignored (S3 signature compatibility).
Value
A Calendar covering the selected part of the year.
Details
A real sample (fewer leaves than the parent) is book-kept exactly like
a geoscales sample: the result is renamed
"base[timeframe:labels-or-hash]" (so two different samples of one
parent never collide in registries or joins), the root parent's name
and totals are recorded in meta$parent_name/meta$parent_totals,
and meta$coverage holds the surviving fraction of share and
weight (read it with calendar_coverage()). Filter-of-filter
composes against the root parent. A filter that keeps everything is a
true no-op: the calendar is returned unchanged.
There is no drop_empty_timeframes= twin of the geoscales argument:
a calendar leaftable is total (no NA memberships), so no timeframe
can empty out under filtering.
cal[timeframe, labels] is subsetting sugar for the same operation.
Examples
win <- filter_calendar(calendar("s4_h24"), "SEASON", "WIN")
win <- calendar("s4_h24")["SEASON", "WIN"] # same
calendar_coverage(win)
#> share weight
#> 0.2465753 0.2465753