Expression-based pipeline

Expression-based pipeline is a group of expressions being evaluated as a pipeline, that is, the whole expression is not evaluated in standard way but regarded as a series of pipeline expressions.

For the following example written with %>>%,

mtcars$mpg %>>%
  sample(size = 10000, replace = TRUE) %>>%
  density(kernel = "gaussian") %>>%
  plot(col = "red", main = "density of mpg (bootstrap)")

we can use pipeline({ ... }) to rewrite it as

pipeline({
  mtcars$mpg
  sample(size = 10000, replace = TRUE)
  density(kernel = "gaussian")
  plot(col = "red", main = "density of mpg (bootstrap)")  
})

Note that this is different from the way we write argument-based pipeline. Instead of supplying the expressions as arguments, we only supply to the first argument an expression enclosed by {} in which each line represents a pipeline step.

Expression-based pipeline differs from argument-based pipeline in that the expression should be quoted by {} and each step must by written in a new line. This is useful because such style, in most cases, makes the code cleaner than without any line breaks.

In addition to the syntactic distinctions, the expression supplied to pipeline() does not require () to enclose task expressions involving symbols that are specially defined in local scope like ~ and ?. In the following example,

mtcars$mpg %>>%
  sample(size = 10000, replace = TRUE) %>>%
  (? length(.)) %>>%
  (~ mtcars_sample) %>>%
  density(kernel = "gaussian") %>>%
  plot(col = "red", main = "density of mpg (bootstrap)")

Both ? length(.) and ~ mtcars_sample should be enclosed by () because of operator priority issues. In pipeline() expressions, such issues do not arise. Therefore, we can perform such special pipeline tasks without enclosing the expression by ().

pipeline({
  mtcars$mpg
  sample(size = 10000, replace = TRUE)
  ? length(.)
  ~ mtcars_sample
  density(kernel = "gaussian")
  plot(col = "red", main = "density of mpg (bootstrap)")  
})

All other features works in pipeline() too. For example, we can ask labeled question in pipeline.

pipeline({
  mtcars
  "Total number of records" ? nrow(.)
  subset(mpg >= quantile(mpg, 0.05) & mpg <= quantile(mpg, 0.95))
  "Qualified number of records" ? nrow(.)
  lm(formula = mpg ~ wt + cyl)
  summary
  "R Squared" ? .$r.squared
  coef
})

# ? Total number of records 
# [1] 32
# ? Qualified number of records 
# [1] 28
# ? R Squared 
# [1] 0.8252262

#              Estimate Std. Error   t value     Pr(>|t|)
# (Intercept) 36.630834  1.6127431 22.713372 3.299463e-18
# wt          -2.528175  0.7657771 -3.301450 2.894825e-03
# cyl         -1.418216  0.3533452 -4.013684 4.783302e-04