Recast data indexed by a product of scales

Converts one or more axes of a product in a single call. Axes named in to are converted; axes left out pass through untouched.

recast_product(
  data,
  x,
  to,
  values = NULL,
  rules = NULL,
  weights = NULL,
  na_action = c("drop", "error", "keep"),
  collect = NULL
)

Arguments

data

The data, in any supported backend, carrying one key column per axis (see product_keys()) plus the value columns.

x

A ScaleProduct.

to

Named list or character vector, one entry per axis to convert: either a frame name of that axis's scale, or another NestedScale to convert into. Axes omitted pass through unchanged.

values

Value columns to convert. Default: all numeric columns that are not key columns. Numeric identifiers (a year) must be excluded explicitly.

rules

Rule per value column and axis: a scalar for everything, or a named list whose entries are scalars or vectors named by axis, e.g. list(load = c(time = "weighted_mean", space = "sum")). NULL (default) resolves each column on each axis through that dimension's own scope in the rule registry – which is where an intensive/extensive asymmetry between dimensions belongs.

weights

Weight column per axis, e.g. c(space = "pop"); scalar applies to every axis.

na_action

Passed to each axis's pass.

collect

For lazy inputs: materialise (TRUE) or return the query.

Value

The recast data in the input's class, keyed by the converted axes' target units.

Details

Execution is a sequence of ordinary single-axis recast_scale() calls – see the file comment in R/product-recast.R for why that is exactly equivalent to a joint recast for every supported rule, and why the joint crosswalk is never built. The passes are ordered so the axis that reduces the data most runs first.

A per-axis "sd" is refused because it is order-dependent; a scalar "sd" means the pooled sd over all product atoms in a target block, which is well defined and is computed without materialising the product. "share" is not defined across a product, because it changes which axis the result is keyed by.

Examples

p <- scale_product(a = scale_example(), b = scale_example2())
d <- merge(
  data.frame(unit = c("U1", "U2", "U3", "U4", "U5", "U6")),
  data.frame(period = c("p1", "p2", "p3", "p4"))
)
d$cap <- seq_len(nrow(d))

# aggregate both axes at once
recast_product(d, p, to = list(a = "sector", b = "era"), rules = "sum")
#>   era sector cap
#> 1  E1      P  44
#> 2  E2      P 140
#> 3  E1      S  34
#> 4  E2      S  82

# different rules per axis
recast_product(d, p,
  to = list(a = "sector", b = "era"),
  rules = list(cap = c(a = "sum", b = "weighted_mean"))
)
#>   era sector cap
#> 1  E1      P  28
#> 2  E2      P  74
#> 3  E1      S  20
#> 4  E2      S  43