relocate

[D-Se]

Relocate executes and yields data table directly, without a as.data.table call

library(data.table)
library(dtplyr)

flights_DT <- dtplyr::lazy_dt(nycflights13::flights) %>%
    as.data.table()

flights_DT %>%
    relocate(where(is.character), .before = where(is.numeric))

[myoung3]

Thanks for reporting.

From what I understand, piping flights_DT into tidyverse queries won't make use of dtplyr functions at all since flights_DT is a data.table and thus will be treated as a data.frame using normal dplyr methods. If you want to use the speed of data.table/dtplyr you need to pipe an object of class dtplyr_step (i.e., the objects that get created by calling lazy_dt on data.tables)

Your code:

flights_DT <- dtplyr::lazy_dt(nycflights13::flights) %>%
    as.data.table()

is just a complicated way of writing as.data.table(nycflights13::flights) and thus your example isn't really using dtplyr at all.

When piping a lazy dt (ie an object of class dtplyr_step) to relocate, the return is also a lazy dt, as is intended.

library(data.table)
library(dplyr,warn.conflicts = FALSE)
library(dtplyr)
flights_lazyDT <- dtplyr::lazy_dt(nycflights13::flights)
flights_DT <- flights_lazyDT %>% as.data.table()

out <- flights_lazyDT %>%
  relocate(where(is.character), .before = where(is.numeric)) 

class(flights_DT)
#> [1] "data.table" "data.frame"
class(out)
#> [1] "dtplyr_step_first" "dtplyr_step"

^{Created on 2021-03-27 by the reprex package (v1.0.0)}

With that said, it does seem like you're found an issue with dplyr and not dtplyr, since it seems the default behavior should be conversion to tibble but that's not happening with relocate:

library(data.table)
library(dplyr,warn.conflicts = FALSE)
flights_DT <- as.data.table(nycflights13::flights)

out2 <- flights_DT %>%
  relocate(where(is.character), .before = where(is.numeric)) 


out3 <- flights_DT %>% 
  group_by(month) %>%
  summarize(x=mean(dep_delay,na.rm=TRUE))

class(out2)
#> [1] "data.table" "data.frame"
class(out3)
#> [1] "tbl_df"     "tbl"        "data.frame"

^{Created on 2021-03-27 by the reprex package (v1.0.0)}

[markfairbanks]

@myoung3 As of dtplyr v1.1.0 one of the new features is that the .data.table method of functions automatically convert to a lazy data.table. For example with mutate():

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

test_dt <- data.table(x = 1:3, y = 1:3)

test_dt %>%
  mutate(double_x = x * 2)
#> Source: local data table [3 x 3]
#> Call:   copy(`_DT1`)[, `:=`(double_x = x * 2)]
#> 
#>       x     y double_x
#>   <int> <int>    <dbl>
#> 1     1     1        2
#> 2     2     2        4
#> 3     3     3        6
#> 
#> # Use as.data.table()/as.data.frame()/as_tibble() to access results

It looks like relocate() doesn't have this implemented (even though it should)

[myoung3]

@markfairbanks Very cool! I had missed that. Sorry, @D-Se. My mistake.

And on the second point I was mistaken as well, it seems dplyr's behavior (whether it returns a tibble or a data.frame) actually depends on the function. https://stackoverflow.com/questions/61067989/which-tidyverse-functions-return-tibbles