The fastest delimited reader for R, 952.97 MB/sec/sec.
But that’s impossible! How can it be so fast?
vroom doesn’t stop to actually read all of your data, it simply indexes where each record is located so it can be read later. The vectors returned use the Altrep framework to lazily load the data on-demand when it is accessed, so you only pay for what you use. This lazy access is done automatically, so no changes to your R data-manipulation code are needed.
vroom also uses multiple threads for indexing, materializing non-character columns, and when writing to further improve performance.
package | version | time (sec) | speedup | throughput |
---|---|---|---|---|
vroom | 1.0.1 | 1.75 | 44.38 | 952.97 MB/sec |
data.table | 1.12.2 | 14.79 | 5.25 | 112.68 MB/sec |
readr | 1.3.1 | 31.49 | 2.46 | 52.93 MB/sec |
read.delim | 3.5.3 | 77.62 | 1.00 | 21.47 MB/sec |
vroom has nearly all of the parsing features of readr for delimited and fixed width files, including
- delimiter guessing*
- custom delimiters (including multi-byte* and Unicode* delimiters)
- specification of column types (including type guessing)
- numeric types (double, integer, number)
- logical types
- datetime types (datetime, date, time)
- categorical types (characters, factors)
- column selection, like
dplyr::select()
* - skipping headers, comments and blank lines
- quoted fields
- double and backslashed escapes
- whitespace trimming
- windows newlines
- reading from multiple files or connections*
- embedded newlines in headers and fields**
- writing delimited files with as-needed quoting.
- robust to invalid inputs (vroom has been extensively tested with the afl fuzz tester)*.
* these are additional features only in vroom.
** requires num_threads = 1
.
Install vroom from CRAN with:
install.packages("vroom")
Alternatively, if you need the development version from GitHub install it with:
# install.packages("devtools")
devtools::install_github("r-lib/vroom")
See getting started to jump start your use of vroom!
vroom uses the same interface as readr to specify column types.
vroom::vroom("mtcars.tsv",
col_types = list(cyl = "i", gear = "f",hp = "i", disp = "_",
drat = "_", vs = "l", am = "l", carb = "i")
)
#> # A tibble: 32 x 10
#> model mpg cyl hp wt qsec vs am gear carb
#> <chr> <dbl> <int> <int> <dbl> <dbl> <lgl> <lgl> <fct> <int>
#> 1 Mazda RX4 21 6 110 2.62 16.5 FALSE TRUE 4 4
#> 2 Mazda RX4 Wag 21 6 110 2.88 17.0 FALSE TRUE 4 4
#> 3 Datsun 710 22.8 4 93 2.32 18.6 TRUE TRUE 4 1
#> # … with 29 more rows
vroom natively supports reading from multiple files (or even multiple connections!).
First we generate some files to read by splitting the nycflights dataset by airline.
library(nycflights13)
purrr::iwalk(
split(flights, flights$carrier),
~ vroom::vroom_write(.x, glue::glue("flights_{.y}.tsv"), delim = "\t")
)
Then we can efficiently read them into one tibble by passing the filenames directly to vroom.
files <- fs::dir_ls(glob = "flights*tsv")
files
#> flights_9E.tsv flights_AA.tsv flights_AS.tsv flights_B6.tsv flights_DL.tsv
#> flights_EV.tsv flights_F9.tsv flights_FL.tsv flights_HA.tsv flights_MQ.tsv
#> flights_OO.tsv flights_UA.tsv flights_US.tsv flights_VX.tsv flights_WN.tsv
#> flights_YV.tsv
vroom::vroom(files)
#> Observations: 336,776
#> Variables: 19
#> chr [ 4]: carrier, tailnum, origin, dest
#> dbl [14]: year, month, day, dep_time, sched_dep_time, dep_delay, arr_time, sched_arr...
#> dttm [ 1]: time_hour
#>
#> Call `spec()` for a copy-pastable column specification
#> Specify the column types with `col_types` to quiet this message
#> # A tibble: 336,776 x 19
#> year month day dep_time sched_dep_time dep_delay arr_time
#> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 2013 1 1 810 810 0 1048
#> 2 2013 1 1 1451 1500 -9 1634
#> 3 2013 1 1 1452 1455 -3 1637
#> # … with 3.368e+05 more rows, and 12 more variables: sched_arr_time <dbl>,
#> # arr_delay <dbl>, carrier <chr>, flight <dbl>, tailnum <chr>,
#> # origin <chr>, dest <chr>, air_time <dbl>, distance <dbl>, hour <dbl>,
#> # minute <dbl>, time_hour <dttm>
The speed quoted above is from a real dataset with 14,776,615 rows and 11 columns, see the benchmark article for full details of the dataset and bench/ for the code used to retrieve the data and perform the benchmarks.
In addition to the arguments to the vroom()
function, you can control
the behavior of vroom with a few environment variables. Generally these
will not need to be set by most users.
VROOM_TEMP_PATH
- Path to the directory used to store temporary files when reading from a R connection. If unset defaults to the R session’s temporary directory (tempdir()
).VROOM_THREADS
- The number of processor threads to use when indexing and parsing. If unset defaults toparallel::detectCores()
.VROOM_SHOW_PROGRESS
- Whether to show the progress bar when indexing. Regardless of this setting the progress bar is disabled in non-interactive settings, R notebooks, when running tests with testthat and when knitting documents.VROOM_CONNECTION_SIZE
- The size (in bytes) of the connection buffer when reading from connections (default is 128 KiB).VROOM_WRITE_BUFFER_LINES
- The number of lines to use for each buffer when writing files (default: 1000).
There are also a family of variables to control use of the Altrep
framework. For versions of R where the Altrep framework is unavailable
(R < 3.5.0) they are automatically turned off and the variables have no
effect. The variables can take one of true
, false
, TRUE
, FALSE
,
1
, or 0
.
VROOM_USE_ALTREP_NUMERICS
- If set use Altrep for all numeric types (defaultfalse
).
There are also individual variables for each type. Currently only
VROOM_USE_ALTREP_CHR
defaults to true
.
VROOM_USE_ALTREP_CHR
VROOM_USE_ALTREP_FCT
VROOM_USE_ALTREP_INT
VROOM_USE_ALTREP_DBL
VROOM_USE_ALTREP_NUM
VROOM_USE_ALTREP_LGL
VROOM_USE_ALTREP_DTTM
VROOM_USE_ALTREP_DATE
VROOM_USE_ALTREP_TIME
RStudio’s environment pane calls object.size()
when it refreshes the
pane, which for Altrep objects can be extremely slow. RStudio 1.2.1335+
includes the fixes
(RStudio#4210,
RStudio#4292) for this
issue, so so it is recommended you use at least that version.
- Gabe Becker, Luke Tierney and Tomas Kalibera for conceiving, Implementing and maintaining the Altrep framework
- Romain François, whose Altrepisode package and related blog-posts were a great guide for creating new Altrep objects in C++.
- Matt Dowle and the rest of the
Rdatatable team,
data.table::fread()
is blazing fast and great motivation!