fsum: Fast (Grouped, Weighted) Sum for Matrix-Like Objects

Description

fsum is a generic function that computes the (column-wise) sum of all values in x, (optionally) grouped by g and/or weighted by w (e.g. to calculate survey totals). The TRA argument can further be used to transform x using its (grouped, weighted) sum.

Usage

fsum(x, ...)
# S3 method for default
fsum(x, g = NULL, w = NULL, TRA = NULL, na.rm = .op[["na.rm"]],
     use.g.names = TRUE, fill = FALSE, nthreads = .op[["nthreads"]], ...)
# S3 method for matrix
fsum(x, g = NULL, w = NULL, TRA = NULL, na.rm = .op[["na.rm"]],
     use.g.names = TRUE, drop = TRUE, fill = FALSE, nthreads = .op[["nthreads"]], ...)
# S3 method for data.frame
fsum(x, g = NULL, w = NULL, TRA = NULL, na.rm = .op[["na.rm"]],
     use.g.names = TRUE, drop = TRUE, fill = FALSE, nthreads = .op[["nthreads"]], ...)
# S3 method for grouped_df
fsum(x, w = NULL, TRA = NULL, na.rm = .op[["na.rm"]],
     use.g.names = FALSE, keep.group_vars = TRUE, keep.w = TRUE, stub = .op[["stub"]],
     fill = FALSE, nthreads = .op[["nthreads"]], ...)

Value

The (w weighted) sum of x, grouped by g, or (if TRA is used) x transformed by its (grouped, weighted) sum.

Arguments

x: a numeric vector, matrix, data frame or grouped data frame (class 'grouped_df').
g: a factor, GRP object, atomic vector (internally converted to factor) or a list of vectors / factors (internally converted to a GRP object) used to group x.
w: a numeric vector of (non-negative) weights, may contain missing values.
TRA: an integer or quoted operator indicating the transformation to perform: 0 - "na" | 1 - "fill" | 2 - "replace" | 3 - "-" | 4 - "-+" | 5 - "/" | 6 - "%" | 7 - "+" | 8 - "*" | 9 - "%%" | 10 - "-%%". See TRA.
na.rm: logical. Skip missing values in x. Defaults to TRUE and implemented at very little computational cost. If na.rm = FALSE a NA is returned when encountered.
use.g.names: logical. Make group-names and add to the result as names (default method) or row-names (matrix and data frame methods). No row-names are generated for data.table's.
fill: logical. Initialize result with 0 instead of NA when na.rm = TRUE e.g. fsum(NA, fill = TRUE) returns 0 instead of NA.
nthreads: integer. The number of threads to utilize. See Details.
drop: matrix and data.frame method: Logical. TRUE drops dimensions and returns an atomic vector if g = NULL and TRA = NULL.
keep.group_vars: grouped_df method: Logical. FALSE removes grouping variables after computation.
keep.w: grouped_df method: Logical. Retain summed weighting variable after computation (if contained in grouped_df).
stub: character. If keep.w = TRUE and stub = TRUE (default), the summed weights column is prefixed by "sum.". Users can specify a different prefix through this argument, or set it to FALSE to avoid prefixing.
...: arguments to be passed to or from other methods. If TRA is used, passing set = TRUE will transform data by reference and return the result invisibly.

Details

The weighted sum (e.g. survey total) is computed as sum(x * w), but in one pass and about twice as efficient. If na.rm = TRUE, missing values will be removed from both x and w i.e. utilizing only x[complete.cases(x,w)] and w[complete.cases(x,w)].

This all seamlessly generalizes to grouped computations, which are performed in a single pass (without splitting the data) and are therefore extremely fast. See Benchmark and Examples below.

When applied to data frames with groups or drop = FALSE, fsum preserves all column attributes. The attributes of the data frame itself are also preserved.

Since v1.6.0 fsum explicitly supports integers. Integers are summed using the long long type in C which is bounded at +-9,223,372,036,854,775,807 (so ~4.3 billion times greater than the minimum/maximum R integer bounded at +-2,147,483,647). If the value of the sum is outside +-2,147,483,647, a double containing the result is returned, otherwise an integer is returned. With groups, an integer results vector is initialized, and an integer overflow error is provided if the sum in any group is outside +-2,147,483,647. Data needs to be coerced to double beforehand in such cases.

Multithreading, added in v1.8.0, applies at the column-level unless g = NULL and nthreads > NCOL(x). Parallelism over groups is not available because sums are computed simultaneously within each group. nthreads = 1L uses a serial version of the code, not parallel code running on one thread. This serial code is always used with less than 100,000 obs (length(x) < 100000 for vectors and matrices), because parallel execution itself has some overhead.

Examples

Run this code

## default vector method
mpg <- mtcars$mpg
fsum(mpg)                         # Simple sum
fsum(mpg, w = mtcars$hp)          # Weighted sum (total): Weighted by hp
fsum(mpg, TRA = "%")              # Simple transformation: obtain percentages of mpg
fsum(mpg, mtcars$cyl)             # Grouped sum
fsum(mpg, mtcars$cyl, mtcars$hp)  # Weighted grouped sum (total)
fsum(mpg, mtcars[c(2,8:9)])       # More groups..
g <- GRP(mtcars, ~ cyl + vs + am) # Precomputing groups gives more speed !
fsum(mpg, g)
fmean(mpg, g) == fsum(mpg, g) / fnobs(mpg, g)
fsum(mpg, g, TRA = "%")           # Percentages by group

## data.frame method
fsum(mtcars)
fsum(mtcars, TRA = "%")
fsum(mtcars, g)
fsum(mtcars, g, TRA = "%")

## matrix method
m <- qM(mtcars)
fsum(m)
fsum(m, TRA = "%")
fsum(m, g)
fsum(m, g, TRA = "%")

## method for grouped data frames - created with dplyr::group_by or fgroup_by
mtcars |> fgroup_by(cyl,vs,am) |> fsum(hp)  # Weighted grouped sum (total)
mtcars |> fgroup_by(cyl,vs,am) |> fsum(TRA = "%")
mtcars |> fgroup_by(cyl,vs,am) |> fselect(mpg) |> fsum()

 
## This compares fsum with data.table and base::rowsum
# Starting with small data
library(data.table)
opts <- set_collapse(nthreads = getDTthreads())
mtcDT <- qDT(mtcars)
f <- qF(mtcars$cyl)

library(microbenchmark)
microbenchmark(mtcDT[, lapply(.SD, sum), by = f],
               rowsum(mtcDT, f, reorder = FALSE),
               fsum(mtcDT, f, na.rm = FALSE), unit = "relative")

# Now larger data
tdata <- qDT(replicate(100, rnorm(1e5), simplify = FALSE)) # 100 columns with 100.000 obs
f <- qF(sample.int(1e4, 1e5, TRUE))                        # A factor with 10.000 groups

microbenchmark(tdata[, lapply(.SD, sum), by = f],
               rowsum(tdata, f, reorder = FALSE),
               fsum(tdata, f, na.rm = FALSE), unit = "relative")
# Reset options
set_collapse(opts)