Join data frame-like objects using an extended DT[i]-style interface to data.table

Write (and optionally run) data.table code for a join using a generalisation of DT[i] syntax with extended arguments and enhanced behaviour. Accepts any data.frame-like inputs (not only data.tables), permits left, right, inner, and full joins, prevents unwanted matches on NA and NaN by default, does not garble join columns in non-equality joins, allows mult on both sides of the join, creates an optional join indicator column, allows specifying which columns to select from each input, and provides convenience options to control column order and prefixing.

If run, the join returns a data.frame, data.table, tibble, sf, or sf-tibble according to context. The generated data.table code can be printed to the console instead of (or as well as) being executed. This feature extends to mock joins, where no inputs are provided, and template code is produced.

dtjoin is the workhorse function for fjoin_inner, fjoin_left, fjoin_right, and fjoin_full, which are wrappers providing a more conventional interface for join operations. These functions are recommended over dtjoin for most users and cases.

Usage

dtjoin(
  .DT = NULL,
  .i = NULL,
  on,
  match.na = FALSE,
  mult = "all",
  mult.DT = "all",
  nomatch = NA,
  nomatch.DT = NULL,
  indicate = FALSE,
  select = NULL,
  select.DT = NULL,
  select.i = NULL,
  both = FALSE,
  on.first = FALSE,
  i.home = FALSE,
  i.first = i.home,
  prefix = if (i.home) "x." else "i.",
  i.class = i.home,
  do = !(is.null(.DT) && is.null(.i)),
  show = !do,
  verbose = FALSE,
  ...
)

Arguments

.DT, .i: data.frame-like objects (plain, data.table, tibble, sf, list, etc.), or else both omitted for a mock join statement with no data.
on: A character vector of join predicates, e.g. c("id", "col_DT == col_i", "date < date", "cost <= budget").
match.na: If TRUE, allow equality matches between NAs or NaNs. Default FALSE.
mult: (as in [.data.table) When a row of .i has multiple matching rows in .DT, which to accept. One of "all" (the default), "first", or "last".
mult.DT: Like mult, but with the roles of .DT and .i reversed, i.e. when a row of .DT has multiple matching rows in .i, which to accept (default "all"). Can be combined with mult. See Details.
nomatch: (as in [.data.table) Either NA (the default) to retain rows of .i with no match in .DT, or NULL to exclude them.
nomatch.DT: Like nomatch but with the roles of .DT and .i reversed, and a different default: either NA to append rows of .DT with no match in .i, or NULL (the default) to leave them out.
indicate: Whether to add a column ".join" at the front of the result, with values 1L if from the "home" table only, 2L if from the "foreign" table only, and 3L if joined from both tables (c.f. _merge in Stata). Default FALSE.
select, select.DT, select.i: Character vectors of columns to be selected from either input if present (select) or specifically from one or other (select.DT, select.i). NULL (the default) selects all columns. Use "" or NA to select no columns. Join columns are always selected. See Details.
both: Whether to include equality join columns from the "foreign" table separately in the output, instead of combining them with those from the "home" table. Default FALSE. Note that non-equality join columns from the foreign table are always included separately.
on.first: Whether to place the join columns from both inputs first in the join result. Default FALSE.
i.home: Whether to treat .i as the "home" table and .DT as the "foreign" table for column prefixing and indicate. Default FALSE, i.e. .DT is the "home" table, as in [.data.table.
i.first: Whether to place .i's columns before .DT's in the join result. The default is to use the value of i.home, i.e. bring .i's columns to the front if .i is the "home" table.
prefix: A prefix to attach to column names in the "foreign" table that are the same as a column name in the "home" table. The default is "i." if the "foreign" table is .i (i.home is FALSE) and "x." if it is .DT (i.home is TRUE).
i.class: Whether the class of the output should be based on .i instead of .DT. The default follows i.home (default FALSE). See Details for how output class and other attributes are set.
do: Whether to execute the join. Default is TRUE unless .DT and .i are both omitted/NULL, in which case a mock join statement is produced.
show: Whether to print the code for the join to the console. Default is the opposite of do. If .DT and .i are both omitted/NULL, mock join code is displayed.
verbose: (passed to [.data.table) Whether data.table should print information to the console during execution. Default FALSE.
...: Further arguments (for internal use).

Value

A data.frame, data.table, (grouped) tibble, sf, or sf-tibble, or else NULL if do is FALSE. See Details.

Details

Input and output class

Each input can be any object with class data.frame, or a plain list of same-length vectors.

The output class depends on .DT by default (but .i with i.class = TRUE) and is as follows:

a data.table if the input is a pure data.table
a tibble if it is a tibble (and a grouped tibble if it has class grouped_df)
an sf if it is an sf with its active geometry selected in the join
a plain data.frame in all other cases

The following attributes are carried through and refreshed: data.table key, tibble groups, sf agr (and bbox etc. of all individual sfc-class columns regardless of output class). See below for specifics. Other classes and attributes are not carried through.

Using `select`, `select.DT`, and `select.i`

Used on its own, select keeps the join columns plus the specified non-join columns from both inputs if present.

If select.DT is provided (and similarly for select.i) then:

if select is also specified, non-join columns of .DT named in either select or select.DT are included
if select is not specified, only non-join columns named in select.DT are included from .DT. Thus e.g. select.DT = "" excludes all of .DT's non-join columns.

Non-existent column names are ignored without warning.

Column order

When select is specified but select.DT and select.i are not, the output consists of all join columns followed by the selected non-join columns from either input in the order given in select.

In all other cases:

columns from .DT come before columns from .i by default (but vice versa if i.first is TRUE)
within each group of columns, non-join columns are in the order given by select.DT/select.i, or in their original data order if no selection is provided
if on.first is TRUE, join columns from both inputs are moved to the front of the overall output.

Using `mult` and `mult.DT`

If both of these arguments are not the default "all", mult is applied first (typically by passing directly to [.data.table) and mult.DT is applied subsequently to eliminate all but the first or last occurrence of each row of .DT from the inner part of the join, producing a 1:1 result. This order of operations can affect the identity of the rows in the inner join.

Displaying code and 'mock joins'

The option of displaying the join code with show = TRUE or by passing null inputs is aimed at data.table users wanting to use the package as a cookbook of recipes for adaptation. If .DT and .i are both NULL, template code is displayed based on join column names implied by on, plus sample non-join column names. select arguments are ignored in this case.

The code displayed is for the join operation after casting the inputs as data.tables if necessary, and before casting the result as a tibble and/or sf if applicable. Note that fjoin departs from the usual j = list() idiom in order to avoid a deep copy of the output made by as.data.table.list. (Likewise, internally it takes only shallow copies of columns when casting inputs or outputs to different classes.)

tibble `groups`

If the relevant input is a grouped tibble (class grouped_df), the output is grouped by the grouping columns that are selected in the result.

data.table `key`s

If .i is a keyed data.table and the output is also a data.table, it inherits .i's key provided nomatch.DT is NULL (i.e. the non-matching rows of .DT are not included in the result). This differs from a data.table DT[i] join, in which the output inherits the key of DT provided it remains sorted on those columns. If not all of the key columns are selected in the result, the leading subset is used.

sf objects and `sfc`-class columns

Joins between two sf objects are supported. The relation-to-geometry attribute agr is inherited from the input supplying the active geometry. All sfc-class columns in the output are refreshed after joining (using sf::st_sfc() with recompute_bbox = TRUE); this is true regardless of whether or not the inputs and output are sfs.

Examples

# An illustration showing:
# - two calls to fjoin_left() (commented out), differing in the `order` argument
# - the resulting calls to dtjoin(), plus `show = TRUE`
# - the generated data.table code and output

# data frames
set.seed(1)
df_x <- data.frame(id_x = 1:3, col_x = paste0("x", 1:3), val = runif(3))
df_y <- data.frame(id_y = rep(4:2, each = 2), col_y = paste0("y", 1:6), val = runif(6))

NULL # section break
#> NULL

# (1) fjoin_left(df_x, df_y, on = "id_x == id_y", mult.x = "first")
dtjoin(
  df_y,
  df_x,
  on = "id_y == id_x",
  mult = "first",
  i.home = TRUE,
  prefix = "R.",
  show = TRUE
)
#> .DT : df_y (cast as data.table)
#> .i  : df_x (cast as data.table)
#> Join: .DT[.i, on = "id_y == id_x", mult = "first", data.frame(id_x, col_x, val = i.val, col_y, R.val = val)]
#> 
#>   id_x col_x       val col_y     R.val
#> 1    1    x1 0.2655087  <NA>        NA
#> 2    2    x2 0.3721239    y5 0.6607978
#> 3    3    x3 0.5728534    y3 0.8983897

# (2) fjoin_left(df_x, df_y, on = "id_x == id_y", mult.x = "first", order = "right")
dtjoin(
  df_x,
  df_y,
  on = "id_x == id_y",
  mult.DT = "first",
  nomatch = NULL,
  nomatch.DT = NA,
  prefix = "R.",
  show = TRUE
)
#> .DT : df_x (cast as data.table)
#> .i  : df_y (cast as data.table)
#> Join: setDF(with(list(fjoin.temp = setDT(setDT(.i[, fjoin.which.i := .I][.DT[, fjoin.which.DT := .I], on = "id_y == id_x", nomatch = NULL, mult = "first", data.frame(id_x = i.id_x, col_x = i.col_x, val = i.val, fjoin.which.DT = i.fjoin.which.DT, fjoin.which.i)])[.i, on = "fjoin.which.i", nomatch = NULL, data.frame(id_x = id_y, col_x, val, fjoin.which.DT, col_y, R.val = i.val)])), rbind(fjoin.temp, setDT(.DT[!fjoin.temp$fjoin.which.DT, data.frame(id_x, col_x, val, fjoin.which.DT)]), fill = TRUE))[, fjoin.which.DT := NULL])[]
#> 
#>   id_x col_x       val col_y     R.val
#> 1    3    x3 0.5728534    y3 0.8983897
#> 2    2    x2 0.3721239    y5 0.6607978
#> 3    1    x1 0.2655087  <NA>        NA

Join data frame-like objects using an extended `DT[i]`-style interface to data.table

Usage

Arguments

Value

Details

Input and output class

Using `select`, `select.DT`, and `select.i`

Column order

Using `mult` and `mult.DT`

Displaying code and 'mock joins'

tibble `groups`

data.table `key`s

sf objects and `sfc`-class columns

See also

Examples

Join data frame-like objects using an extended DT[i]-style interface to data.table

Usage

Arguments

Value

Details

Input and output class

Using select, select.DT, and select.i

Column order

Using mult and mult.DT

Displaying code and 'mock joins'

tibble groups

data.table keys

sf objects and sfc-class columns

See also

Examples

Join data frame-like objects using an extended `DT[i]`-style interface to data.table

Using `select`, `select.DT`, and `select.i`

Using `mult` and `mult.DT`

tibble `groups`

data.table `key`s

sf objects and `sfc`-class columns