Translation

Introduction

This vignette shows the details of how dtplyr translates dplyr expressions into the equivalent data.table code. If you see places where you think I could generate better data.table code, please let me know!

This document assumes that you’re familiar with the basics of data.table; if you’re not, I recommend starting at vignette("datatable-intro.html").

library(dtplyr)
library(data.table)
library(dplyr)

The basics

To get started, I’ll create a simple lazy table with lazy_dt():

df <- data.frame(a = 1:5, b = 1:5, c = 1:5, d = 1:5)
dt <- lazy_dt(df)

The actual data doesn’t matter here since we’re just looking at the translation.

When you print a lazy frame, it tells you that it’s a local data table with four rows. It also prints the call that dtplyr will evaluate when we execute the lazy table. In this case it’s very simple:

dt
#> Source: local data table [5 x 4]
#> Call:   `_DT1`
#> 
#>       a     b     c     d
#>   <int> <int> <int> <int>
#> 1     1     1     1     1
#> 2     2     2     2     2
#> 3     3     3     3     3
#> 4     4     4     4     4
#> 5     5     5     5     5
#> 
#> # Use as.data.table()/as.data.frame()/as_tibble() to access results

If we just want to see the generated code, you can use show_query(). I’ll use that a lot in this vignette.

dt %>% show_query()
#> `_DT1`

Simple verbs

Many dplyr verbs have a straightforward translation to either the i or j component of [.data.table.

dt %>% arrange(a, b, c) %>% show_query()
#> `_DT1`[order(a, b, c)]

dt %>% filter(b == c) %>% show_query()
#> `_DT1`[b == c]
dt %>% filter(b == c, c == d) %>% show_query()
#> `_DT1`[b == c & c == d]

dt %>% select(a:b) %>% show_query()
#> `_DT1`[, .(a, b)]
dt %>% summarise(a = mean(a)) %>% show_query()
#> `_DT1`[, .(a = mean(a))]
dt %>% transmute(a2 = a * 2) %>% show_query()
#> `_DT1`[, .(a2 = a * 2)]

dt %>% mutate(a2 = a * 2, b2 = b * 2) %>% show_query()
#> copy(`_DT1`)[, `:=`(a2 = a * 2, b2 = b * 2)]

dt %>% mutate(a2 = a * 2, b2 = b * 2, a4 = a2 * 2) %>% show_query()
#> copy(`_DT1`)[, `:=`(c("a2", "b2", "a4"), {
#>     a2 <- a * 2
#>     b2 <- b * 2
#>     a4 <- a2 * 2
#>     .(a2, b2, a4)
#> })]

dt %>% transmute(a2 = a * 2, b2 = b * 2, a4 = a2 * 2) %>% show_query()
#> `_DT1`[, {
#>     a2 <- a * 2
#>     b2 <- b * 2
#>     a4 <- a2 * 2
#>     .(a2, b2, a4)
#> }]

Other calls

Other verbs require calls to other functions:

dt %>% rename(x = a, y = b) %>% show_query()
#> setnames(copy(`_DT1`), c("a", "b"), c("x", "y"))

dt %>% distinct() %>% show_query()
#> unique(`_DT1`)
dt %>% distinct(a, b) %>% show_query()
#> unique(`_DT1`[, .(a, b)])
dt %>% distinct(a, b, .keep_all = TRUE) %>% show_query()
#> unique(`_DT1`, by = c("a", "b"))

dt %>% distinct(c = a + b) %>% show_query()
#> unique(`_DT1`[, .(c = a + b)])
dt %>% distinct(c = a + b, .keep_all = TRUE) %>% show_query()
#> unique(copy(`_DT1`)[, `:=`(c = a + b)], by = "c")

dt2 <- lazy_dt(data.frame(a = 1))

dt %>% inner_join(dt2, by = "a") %>% show_query()
#> `_DT1`[`_DT2`, on = .(a), nomatch = NULL, allow.cartesian = TRUE]
dt %>% right_join(dt2, by = "a") %>% show_query()
#> `_DT1`[`_DT2`, on = .(a), allow.cartesian = TRUE]
dt %>% left_join(dt2, by = "a") %>% show_query()
#> `_DT2`[`_DT1`, on = .(a), allow.cartesian = TRUE]
dt %>% anti_join(dt2, by = "a") %>% show_query()
#> `_DT1`[!`_DT2`, on = .(a)]

dt %>% full_join(dt2, by = "a") %>% show_query()
#> merge(`_DT1`, `_DT2`, all = TRUE, by.x = "a", by.y = "a", allow.cartesian = TRUE)

dt3 <- lazy_dt(data.frame(b = 1, a = 1))

dt %>% left_join(dt3, by = "a") %>% show_query()
#> setnames(setcolorder(`_DT3`[`_DT1`, on = .(a), allow.cartesian = TRUE], 
#>     c(2L, 3L, 4L, 5L, 1L)), c("i.b", "b"), c("b.x", "b.y"))
dt %>% full_join(dt3, by = "b") %>% show_query()
#> setcolorder(merge(`_DT1`, `_DT3`, all = TRUE, by.x = "b", by.y = "b", 
#>     allow.cartesian = TRUE), c(2L, 1L, 3L, 4L, 5L))

dt %>% semi_join(dt2, by = "a") %>% show_query()
#> `_DT1`[unique(`_DT1`[`_DT2`, which = TRUE, nomatch = NULL, on = .(a)])]

dt %>% intersect(dt2) %>% show_query()
#> fintersect(`_DT1`, `_DT2`)
dt %>% setdiff(dt2) %>% show_query()
#> fsetdiff(`_DT1`, `_DT2`)
dt %>% union(dt2) %>% show_query()
#> funion(`_DT1`, `_DT2`)

Grouping

Just like in dplyr, group_by() doesn’t do anything by itself, but instead modifies the operation of downstream verbs. This generally just involves using the keyby argument:

dt %>% group_by(a) %>% summarise(b = mean(b)) %>% show_query()
#> `_DT1`[, .(b = mean(b)), keyby = .(a)]

You may use by instead of keyby if you set arrange = FALSE:

dt %>% group_by(a, arrange = FALSE) %>% summarise(b = mean(b)) %>% show_query()
#> `_DT1`[, .(b = mean(b)), by = .(a)]

Often, there won’t be too much of a difference between these, but for larger grouped operations, the overhead of reordering data may become significant. In these situations, using arrange = FALSE becomes preferable.

The primary exception is grouped filter(), which requires the use of .SD:

dt %>% group_by(a) %>% filter(b < mean(b)) %>% show_query()
#> `_DT1`[`_DT1`[, .I[b < mean(b)], by = .(a)]$V1]

Combinations

dtplyr tries to generate generate data.table code as close as possible to what you’d write by hand, as this tends to unlock data.table’s tremendous speed. For example, if you filter() and then select(), dtplyr generates a single [:

dt %>% 
  filter(a == 1) %>% 
  select(-a) %>% 
  show_query()
#> `_DT1`[a == 1, .(b, c, d)]

And similarly when combining filtering and summarising:

dt %>% 
  group_by(a) %>% 
  filter(b < mean(b)) %>% 
  summarise(c = max(c)) %>% 
  show_query()
#> `_DT1`[`_DT1`[, .I[b < mean(b)], by = .(a)]$V1, .(c = max(c)), 
#>     keyby = .(a)]

This is particularly nice when joining two tables together because you can select variables after you have joined and data.table will only carry those into the join:

dt3 <- lazy_dt(data.frame(x = 1, y = 2))
dt4 <- lazy_dt(data.frame(x = 1, a = 2, b = 3, c = 4, d = 5, e = 7))

dt3 %>% 
  left_join(dt4) %>% 
  select(x, a:c) %>% 
  show_query()
#> Joining, by = "x"
#> setcolorder(`_DT5`[`_DT4`, on = .(x), allow.cartesian = TRUE], 
#>     c(1L, 7L, 2L, 3L, 4L, 5L, 6L))[, `:=`(c("y", "d", "e"), NULL)]

Note, however, that select()ing and then filter()ing must generate two separate calls to [, because data.table evaluates i before j.

dt %>% 
  select(X = a, Y = b) %>% 
  filter(X == 1) %>% 
  show_query()
#> `_DT1`[, .(X = a, Y = b)][X == 1]

Similarly, a filter() and mutate() can’t be combined because dt[a == 1, .(b2 := b * 2)] would modify the selected rows in place:

dt %>% 
  filter(a == 1) %>% 
  mutate(b2 = b * 2) %>% 
  show_query()
#> `_DT1`[a == 1][, `:=`(b2 = b * 2)]

Copies

By default dtplyr avoids mutating the input data, automatically creating a copy() if needed:

dt %>% mutate(a2 = a * 2, b2 = b * 2) %>% show_query()
#> copy(`_DT1`)[, `:=`(a2 = a * 2, b2 = b * 2)]

Note that dtplyr does its best to avoid needless copies, so it won’t explicitly copy if there’s already an implicit copy produced by [, head(), merge() or similar:

dt %>% 
  filter(x == 1) %>% 
  mutate(a2 = a * 2, b2 = b * 2) %>% 
  show_query()
#> `_DT1`[x == 1][, `:=`(a2 = a * 2, b2 = b * 2)]

You can choose to opt out of this copy, and take advantage of data.table’s reference semantics (see vignette("datatable-reference-semantics") for more details). Do this by setting immutable = FALSE on construction:

dt2 <- data.table(a = 1:10)

dt_inplace <- lazy_dt(dt2, immutable = FALSE)
dt_inplace %>% mutate(a2 = a * 2, b2 = b * 2) %>% show_query()
#> `_DT6`[, `:=`(a2 = a * 2, b2 = b * 2)]

Performance

There are two components to the performance of dtplyr: how long it takes to generate the translation, and how well the translation performs. Given my explorations so far, I’m reasonably confident that we’re generating high-quality data.table code, so most of the cost should be in the translation itself.

The following code briefly explores the performance of a few different translations. A significant amount of work is done by the dplyr verbs, so we benchmark the whole process.

bench::mark(
  filter = dt %>% filter(a == b, c == d),
  mutate = dt %>% mutate(a = a * 2, a4 = a2 * 2, a8 = a4 * 2) %>% show_query(),
  summarise = dt %>% group_by(a) %>% summarise(b = mean(b)) %>% show_query(),
  check = FALSE
)[1:6]
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 filter     727.89µs 749.88µs     1296.    3.25KB     10.4
#> 2 mutate     983.93µs   1.02ms      981.   19.73KB     10.4
#> 3 summarise    1.73ms   1.77ms      564.   28.73KB     10.4

These translations all take less than a millisecond, suggesting that the performance overhead of dtplyr should be negligible for realistic data sizes. Note that dtplyr run-time scales with the complexity of the pipeline, not the size of the data, so these timings should apply regardless of the size of the underlying data¹.