Recast values through a given crosswalk — recast

The engine behind recast_scale(), for a crosswalk built elsewhere: by a dimension package whose conversion cannot be read off one scale's leaftable, such as timescales mapping one calendar onto another through a datetime grid. The crosswalk says how much of each source code falls in each target code; the rules then aggregate exactly as in recast_scale(), on any supported backend.

recast_crosswalk(
  data,
  map,
  from,
  to,
  key = from,
  values = NULL,
  rule = NULL,
  ids = NULL,
  by = character(),
  targets = NULL,
  missing_sources = c("na", "ignore"),
  unit = "unit",
  collect = NULL
)

Arguments

data: The data, in any supported backend, with a code column and one or more numeric value columns.
map: The crosswalk, a data.frame.
from, to: Names of the source and target code columns of map.
key: Name of the code column of data. Defaults to from.
values: Value columns. Default: every numeric column that is neither key nor in ids.
rule: One of SCALE_RULES other than the share rules, for every value column, or a named vector with one entry per column. NULL looks each column up with get_scale_rule().
ids: Identifier columns, preserved as groups. Default: every column that is neither key nor a value column. Other columns are dropped.
by: Columns of map matched against the identically named identifier columns of data, for a crosswalk that differs between identifier values (a calendar crosswalk per year, say).
targets: Optional full target vocabulary. A materialised result is completed to every target per identifier combination, in this order, with NA where nothing landed; an NA entry adds an explicit NA row.
missing_sources: "na" (default) or "ignore"; see recast_scale().
unit: The word for a target code in error messages.
collect: For lazy inputs: materialise (TRUE) or return the query.

Value

The recast data in its own class, with columns c(to, ids, values). Lazy in, lazy out.

Details

The crosswalk is a data frame with one row per (source, target) pair:

the source and target codes, in the columns named by from and to;
n_from, the number of atoms (grid points) in the source code, and n_overlap, the number of them that fall in the target;
optionally w, the weight of the pair for "weighted_mean" – by default n_overlap;
optionally w_from, the source's total weight. With it, "sum" splits a source across its targets by w / w_from; without it, by n_overlap / n_from, an equal split over atoms;
any columns named in by.

A target whose sources are only partly present in the data, per identifier combination, comes back NA unless missing_sources = "ignore".

Examples

# Two months onto one quarter, with the months' day counts as atoms
map <- data.frame(
  month = c("m01", "m02"), quarter = "Q1",
  n_from = c(31, 28), n_overlap = c(31, 28)
)
d <- data.frame(
  month = c("m01", "m02"), energy = c(310, 280),
  price = c(10, 20)
)
recast_crosswalk(d, map,
  from = "month", to = "quarter",
  rule = c(energy = "sum", price = "weighted_mean")
)
#>   quarter energy    price
#> 1      Q1    590 14.74576