Compare data.table pipes and magrittr pipes

We have two ways to chain data.table operations, using data.table pipes or using magrittr pipes. For a data.table dt, the data.table pipes take the form of dt[][][]... and magrittr pipes dt %>% .[] %>% .[] %>% ....

Let’s first compare the readability of the two pipes in the following examples. Hadley Wickham criticized the readability of data.table pipes in this stackoverflow post. The data.table pipes, however, are not that hard to follow for those who are familiar with data.table. To my eyes, the magrittr pipes improve the readability but the data.table pipes are still acceptable. It is more of a personal choice.

library(data.table)
library(magrittr)

# the data.table pipes
data.table(iris)[
    # add a new column "is_setosa", 1 if yes and 0 if no, use two lines of codes 
    # to clearly show the values
    Species == "setosa", is_setosa := 1
][
    Species != "setosa", is_setosa := 0
][
    # changes the Petal.Length of Species "versicolor", other species not affected
    Species == "versicolor", Petal.Length := 999
][
    # calculate sepal area when length > 5. Area is NA if length <= 5
    Sepal.Length > 5, Sepal.Area := Sepal.Length * Sepal.Width
][
    # select columns
    , .(Species, is_setosa, Petal.Length, Sepal.Area)
][
    # average of each column grouped by species
    , lapply(.SD, mean, na.rm = TRUE), by = Species
]

##       Species is_setosa Petal.Length Sepal.Area
## 1:     setosa         1        1.462   19.76364
## 2: versicolor         0      999.000   16.87340
## 3:  virginica         0        5.552   19.83633

# the magrittr pipes
data.table(iris) %>%
    # add a new column "is_setosa", 1 if yes and 0 if no, use two lines of codes 
    # to clearly show the values
    .[Species == "setosa", is_setosa := 1] %>%                         
    .[Species != "setosa", is_setosa := 0] %>%
    
    # changes the Petal.Length of Species "versicolor", other species not affected
    .[Species == "versicolor", Petal.Length := 999] %>%   
    
    # calculate sepal area when length > 5. Area is NA if length <= 5
    .[Sepal.Length > 5, Sepal.Area := Sepal.Length * Sepal.Width] %>%  
    
    # select columns
    .[, .(Species, is_setosa, Petal.Length, Sepal.Area)] %>%    
    
    # average of each column grouped by species
    .[, lapply(.SD, mean, na.rm = TRUE), by = Species]

##       Species is_setosa Petal.Length Sepal.Area
## 1:     setosa         1        1.462   19.76364
## 2: versicolor         0      999.000   16.87340
## 3:  virginica         0        5.552   19.83633

Will the use of pipes %>% slow down the computing? In the following code, we add three new columns to a made-up data table with data.table pipes and magrittr pipes. They almost have the same speed.

set.seed(123)
dt <- data.table(a = sample(letters, 1e5, replace = TRUE),
                 b = abs(rnorm(1e5)))

datatable_pipe <- function(){
    dt[, x := sqrt(b)][
        , y := b^2
    ][
        , z := paste0(a , b)
    ]
}

magrittr_pipe <- function(){
    dt[, x := sqrt(b)] %>%
        .[, y := b^2] %>%
        .[, z := paste0(a , b)]
}

rbenchmark::benchmark(datatable_pipe(), magrittr_pipe(), replications=20)

##               test replications elapsed relative user.self sys.self
## 1 datatable_pipe()           20   3.883    1.059     3.869    0.012
## 2  magrittr_pipe()           20   3.667    1.000     3.657    0.008
##   user.child sys.child
## 1          0         0
## 2          0         0