In moodymudskipper/withDT: Use data.table Syntax For One Call

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

withDT

withDT features 2 functions, withDT() and %DT>%`() which make data.table syntax available for single calls without further class or attribute housekeeping.

• withDT() is called to use data.table syntax for one call.
• %DT>% is a pipe that allows to insert data.table calls in magrittr pipe chains.

Some benefits are :

• Leverage compact data.table syntax, and speed to a degree
• Clearly isolate data.table code to avoid confusion due to hybrid syntax
• Be sure that no data.table object is created in calling environment, avoiding potential confusion
• Keep the class of x (regular data frame, tibble or other), with some exceptions detailed below
• data.table is not attached so there is no risk of masking functions from other packages such as lubridate (11 conflicting functions), dplyr (4 conflicting functions) or purrr (1 conflicting function)

We believe this package can help integrating some powerful data.table features in any workflow with minimal confusion.

Installation

Install with:

# install.packages("devtools")
devtools::install_github("moodymudskipper/withDT")

Assignments by reference

An important particularity of withDT() is that assignments are never done by reference. Though limiting it avoids the confusion and unintended behaviors that might come with them.

The syntax of these assignments is still supported but will return a copy. In order to fail explicitly whenever that syntax is used, the argument lock can be set to TRUE the withDT.lock option can be set with options(withDT.lock = TRUE).

Examples

withDT()

library(withDT)

environment(withDT) <- .GlobalEnv

We can use standard data.table syntax and the output will have the same class as the input.

iris2 <- withDT(iris[, .(meanSW = mean(Sepal.Width)), by = Species][,cbind(.SD, a =3)])
iris2
class(iris2)

The following also works, but wouldn't have the same output with standard data.table code due to the way assignments by reference are handled.

However we think in these cases the behavior of withDT() is more likely to be expected.

iris3 <- withDT(iris[, .(meanSW = mean(Sepal.Width)), by = Species][,a:=3])
identical(iris2,iris3)

We see that iris wasn't modified

class(iris)
names(iris)

To trigger an error when this syntax is used we can set lock to TRUE.

iris4 <- withDT(lock=TRUE,iris[, .(meanSW = mean(Sepal.Width)), by = Species][,b:=3])

%DT>%

The %DT>% pipe is another way to use data.table's power and syntax. It can in fact be more efficient than withDT().

It can be used on simple calls :

iris %DT>% .[, .(meanSW = mean(Sepal.Width)), by = Species]

Or be used as part of a pipe chain using magrittr 's operator %>%, for example we can mix data.table and tidyverse operations by doing:

library(tibble)
library(dplyr, warn.conflicts = FALSE)
iris %>%
  as_tibble() %DT>%
  .[, .(meanSW = mean(Sepal.Width)), by = Species][
    ,Species := as.character(Species)] %>%
  filter(startsWith(Species,"v"))

Some caveats

• Copies are created for every call of [, which might be costly in some cases, if optimization is necessary we suggest using data.table::setDT() followed by idiomatic data.table syntax
• Attributes other than class are stripped
• Classes grouped_df and rowwise_df, created respectively by dplyr::group_by() and dplyr::rowwise() are stripped
• When using syntax of assignment by reference and lock is FALSE (default), some expressions won't be equivalent to the data.table code, see examples
• At the moment withDT() fails in markdown unless we assign it to the global environment.

A workaround to the latter is to include the following chunk at the top of your document :

cat("```r", "environment(withDT) <- .GlobalEnv","```", sep = "\n")

moodymudskipper/withDT documentation built on Nov. 4, 2019, 7:29 p.m.