Fast and friendly file finagler
Similar to read.table but faster and more convenient. All controls such as sep, colClasses and nrows are automatically detected.
fread is for regular delimited files; i.e., where every row has the same number of columns. In future, secondary separator (sep2) may be specified within each column. Such columns will be read as type list where each cell is itself a vector.
fread(input, file, text, cmd, sep="auto", sep2="auto", dec=".", quote="\"",
nrows=Inf, header="auto",
na.strings=getOption("datatable.na.strings","NA"), # due to change to ""; see NEWS
stringsAsFactors=FALSE, verbose=getOption("datatable.verbose", FALSE),
skip="__auto__", select=NULL, drop=NULL, colClasses=NULL,
integer64=getOption("datatable.integer64", "integer64"),
col.names,
check.names=FALSE, encoding="unknown",
strip.white=TRUE, fill=FALSE, blank.lines.skip=FALSE,
key=NULL, index=NULL,
showProgress=getOption("datatable.showProgress", interactive()),
data.table=getOption("datatable.fread.datatable", TRUE),
nThread=getDTthreads(verbose),
logical01=getOption("datatable.logical01", FALSE), # due to change to TRUE; see NEWS
keepLeadingZeros = getOption("datatable.keepLeadingZeros", FALSE),
yaml=FALSE, autostart=NA, tmpdir=tempdir(), tz="UTC"
)input |
A single character string. The value is inspected and deferred to either |
file |
File name in working directory, path to file (passed through |
text |
The input data itself as a character vector of one or more lines, for example as returned by |
cmd |
A shell command that pre-processes the file; e.g. |
sep |
The separator between columns. Defaults to the character in the set |
sep2 |
The separator within columns. A |
nrows |
The maximum number of rows to read. Unlike |
header |
Does the first data line contain column names? Defaults according to whether every non-empty field on the first data line is type character. If so, or TRUE is supplied, any empty column names are given a default name. |
na.strings |
A character vector of strings which are to be interpreted as |
stringsAsFactors |
Convert all character columns to factors? |
verbose |
Be chatty and report timings? |
skip |
If 0 (default) start on the first line and from there finds the first row with a consistent number of columns. This automatically avoids irregular header information before the column names row. |
select |
A vector of column names or numbers to keep, drop the rest. |
drop |
Vector of column names or numbers to drop, keep the rest. |
colClasses |
As in |
integer64 |
"integer64" (default) reads columns detected as containing integers larger than 2^31 as type |
dec |
The decimal separator as in |
col.names |
A vector of optional names for the variables (columns). The default is to use the header column if present or detected, or if not "V" followed by the column number. This is applied after |
check.names |
default is |
encoding |
default is |
quote |
By default ( |
strip.white |
default is |
fill |
logical (default is |
blank.lines.skip |
|
key |
Character vector of one or more column names which is passed to |
index |
Character vector or list of character vectors of one or more column names which is passed to |
showProgress |
|
data.table |
TRUE returns a |
nThread |
The number of threads to use. Experiment to see what works best for your data on your hardware. |
logical01 |
If TRUE a column containing only 0s and 1s will be read as logical, otherwise as integer. |
keepLeadingZeros |
If TRUE a column containing numeric data with leading zeros will be read as character, otherwise leading zeros will be removed and converted to numeric. |
yaml |
If |
autostart |
Deprecated and ignored with warning. Please use |
tmpdir |
Directory to use as the |
tz |
Relevant to datetime values which have no Z or UTC-offset at the end, i.e. unmarked datetime, as written by |
A sample of 10,000 rows is used for a very good estimate of column types. 100 contiguous rows are read from 100 equally spaced points throughout the file including the beginning, middle and the very end. This results in a better guess when a column changes type later in the file (e.g. blank at the beginning/only populated near the end, or 001 at the start but 0A0 later on). This very good type guess enables a single allocation of the correct type up front once for speed, memory efficiency and convenience of avoiding the need to set colClasses after an error. Even though the sample is large and jumping over the file, it is almost instant regardless of the size of the file because a lazy on-demand memory map is used. If a jump lands inside a quoted field containing newlines, each newline is tested until 5 lines are found following it with the expected number of fields. The lowest type for each column is chosen from the ordered list: logical, integer, integer64, double, character. Rarely, the file may contain data of a higher type in rows outside the sample (referred to as an out-of-sample type exception). In this event fread will automatically reread just those columns from the beginning so that you don't have the inconvenience of having to set colClasses yourself; particularly helpful if you have a lot of columns. Such columns must be read from the beginning to correctly distinguish "00" from "000" when those have both been interpreted as integer 0 due to the sample but 00A occurs out of sample. Set verbose=TRUE to see a detailed report of the logic deployed to read your file.
There is no line length limit, not even a very large one. Since we are encouraging list columns (i.e. sep2) this has the potential to encourage longer line lengths. So the approach of scanning each line into a buffer first and then rescanning that buffer is not used. There are no buffers used in fread's C code at all. The field width limit is limited by R itself: the maximum width of a character string (currently 2^31-1 bytes, 2GB).
The filename extension (such as .csv) is irrelevant for "auto" sep and sep2. Separator detection is entirely driven by the file contents. This can be useful when loading a set of different files which may not be named consistently, or may not have the extension .csv despite being csv. Some datasets have been collected over many years, one file per day for example. Sometimes the file name format has changed at some point in the past or even the format of the file itself. So the idea is that you can loop fread through a set of files and as long as each file is regular and delimited, fread can read them all. Whether they all stack is another matter but at least each one is read quickly without you needing to vary colClasses in read.table or read.csv.
If an empty line is encountered then reading stops there with warning if any text exists after the empty line such as a footer. The first line of any text discarded is included in the warning message. Unless, it is single-column input. In that case blank lines are significant (even at the very end) and represent NA in the single column. So that fread(fwrite(DT))==DT. This default behaviour can be controlled using blank.lines.skip=TRUE|FALSE.
Line endings: All known line endings are detected automatically: \n (*NIX including Mac), \r\n (Windows CRLF), \r (old Mac) and \n\r (just in case). There is no need to convert input files first. fread running on any architecture will read a file from any architecture. Both \r and \n may be embedded in character strings (including column names) provided the field is quoted.
Decimal separator and locale: fread(...,dec=",") should just work. fread uses C function strtod to read numeric data; e.g., 1.23 or 1,23. strtod retrieves the decimal separator (. or , usually) from the locale of the R session rather than as an argument passed to the strtod function. So for fread(...,dec=",") to work, fread changes this (and only this) R session's locale temporarily to a locale which provides the desired decimal separator.
On Windows, "French_France.1252" is tried which should be available as standard (any locale with comma decimal separator would suffice) and on unix "fr_FR.utf8" (you may need to install this locale on unix). fread() is very careful to set the locale back again afterwards, even if the function fails with an error. The choice of locale is determined by options()$datatable.fread.dec.locale. This may be a vector of locale names and if so they will be tried in turn until the desired dec is obtained; thus allowing more than two different decimal separators to be selected. This is a new feature in v1.9.6 and is experimental. In case of problems, turn it off with options(datatable.fread.dec.experiment=FALSE).
Quotes:
When quote is a single character,
Spaces and other whitespace (other than sep and \n) may appear in unquoted character fields, e.g., ...,2,Joe Bloggs,3.14,....
When character columns are quoted, they must start and end with that quoting character immediately followed by sep or \n, e.g., ...,2,"Joe Bloggs",3.14,....
In essence quoting character fields are required only if sep or \n appears in the string value. Quoting may be used to signify that numeric data should be read as text. Unescaped quotes may be present in a quoted field, e.g., ...,2,"Joe, "Bloggs"",3.14,..., as well as escaped quotes, e.g., ...,2,"Joe \",Bloggs\"",3.14,....
If an embedded quote is followed by the separator inside a quoted field, the embedded quotes up to that point in that field must be balanced; e.g. ...,2,"www.blah?x="one",y="two"",3.14,....
On those fields that do not satisfy these conditions, e.g., fields with unbalanced quotes, fread re-attempts that field as if it isn't quoted. This is quite useful in reading files that contains fields with unbalanced quotes as well, automatically.
To read fields as is instead, use quote = "".
CSVY Support:
Currently, the yaml setting is somewhat inflexible with respect to incorporating metadata to facilitate file reading. Information on column classes should be stored at the top level under the heading schema and subheading fields; those with both a type and a name sub-heading will be merged into colClasses. Other supported elements are as follows:
sep (or alias delimiter)
header
quote (or aliases quoteChar, quote_char)
dec (or alias decimal)
na.strings
File Download:
When input begins with http://, https://, ftp://, ftps://, or file://, fread detects this and downloads the target to a temporary file (at tempfile()) before proceeding to read the file as usual. Secure URLS (ftps:// and https://) are downloaded with curl::curl_download; ftp:// and http:// paths are downloaded with download.file and method set to getOption("download.file.method"), defaulting to "auto"; and file:// is downloaded with download.file with method="internal". NB: this implies that for file://, even files found on the current machine will be "downloaded" (i.e., hard-copied) to a temporary file. See download.file for more details.
Shell commands:
A data.table by default, otherwise a data.frame when argument data.table=FALSE.
Background :
https://cran.r-project.org/doc/manuals/R-data.html
https://stackoverflow.com/questions/1727772/quickly-reading-very-large-tables-as-dataframes-in-r
https://cerebralmastication.com/2009/11/loading-big-data-into-r/
https://stackoverflow.com/questions/9061736/faster-than-scan-with-rcpp
https://stackoverflow.com/questions/415515/how-can-i-read-and-manipulate-csv-file-data-in-c
https://stackoverflow.com/questions/9352887/strategies-for-reading-in-csv-files-in-pieces
https://stackoverflow.com/questions/11782084/reading-in-large-text-files-in-r
https://stackoverflow.com/questions/45972/mmap-vs-reading-blocks
https://stackoverflow.com/questions/258091/when-should-i-use-mmap-for-file-access
https://stackoverflow.com/a/9818473/403310
https://stackoverflow.com/questions/9608950/reading-huge-files-using-memory-mapped-files
finagler = "to get or achieve by guile or manipulation" https://dictionary.reference.com/browse/finagler
On YAML, see https://yaml.org/; on csvy, see https://csvy.org/.
# Reads text input directly :
fread("A,B\n1,2\n3,4")
# Reads pasted input directly :
fread("A,B
1,2
3,4
")
# Finds the first data line automatically :
fread("
This is perhaps a banner line or two or ten.
A,B
1,2
3,4
")
# Detects whether column names are present automatically :
fread("
1,2
3,4
")
# Numerical precision :
DT = fread("A\n1.010203040506070809010203040506\n")
# TODO: add numerals=c("allow.loss", "warn.loss", "no.loss") from base::read.table, +"use.Rmpfr"
typeof(DT$A)=="double" # currently "allow.loss" with no option
DT = fread("A\n1.46761e-313\n") # read as 'numeric'
DT[,sprintf("%.15E",A)] # beyond what double precision can store accurately to 15 digits
# For greater accuracy use colClasses to read as character, then package Rmpfr.
# colClasses
data = "A,B,C,D\n1,3,5,7\n2,4,6,8\n"
fread(data, colClasses=c(B="character",C="character",D="character")) # as read.csv
fread(data, colClasses=list(character=c("B","C","D"))) # saves typing
fread(data, colClasses=list(character=2:4)) # same using column numbers
# drop
fread(data, colClasses=c("B"="NULL","C"="NULL")) # as read.csv
fread(data, colClasses=list(NULL=c("B","C"))) #
fread(data, drop=c("B","C")) # same but less typing, easier to read
fread(data, drop=2:3) # same using column numbers
# select
# (in read.csv you need to work out which to drop)
fread(data, select=c("A","D")) # less typing, easier to read
fread(data, select=c(1,4)) # same using column numbers
# select and types combined
fread(data, select=c(A="numeric", D="character"))
fread(data, select=list(numeric="A", character="D"))
# skip blank lines
fread("a,b\n1,a\n2,b\n\n\n3,c\n", blank.lines.skip=TRUE)
# fill
fread("a,b\n1,a\n2\n3,c\n", fill=TRUE)
fread("a,b\n\n1,a\n2\n\n3,c\n\n", fill=TRUE)
# fill with skip blank lines
fread("a,b\n\n1,a\n2\n\n3,c\n\n", fill=TRUE, blank.lines.skip=TRUE)
# check.names usage
fread("a b,a b\n1,2\n")
fread("a b,a b\n1,2\n", check.names=TRUE) # no duplicates + syntactically valid names
## Not run:
# Demo speed-up
n = 1e6
DT = data.table( a=sample(1:1000,n,replace=TRUE),
b=sample(1:1000,n,replace=TRUE),
c=rnorm(n),
d=sample(c("foo","bar","baz","qux","quux"),n,replace=TRUE),
e=rnorm(n),
f=sample(1:1000,n,replace=TRUE) )
DT[2,b:=NA_integer_]
DT[4,c:=NA_real_]
DT[3,d:=NA_character_]
DT[5,d:=""]
DT[2,e:=+Inf]
DT[3,e:=-Inf]
write.table(DT,"test.csv",sep=",",row.names=FALSE,quote=FALSE)
cat("File size (MB):", round(file.info("test.csv")$size/1024^2),"\n")
# 50 MB (1e6 rows x 6 columns)
system.time(DF1 <-read.csv("test.csv",stringsAsFactors=FALSE))
# 5.4 sec (first time in fresh R session)
system.time(DF1 <- read.csv("test.csv",stringsAsFactors=FALSE))
# 3.9 sec (immediate repeat is faster, varies)
system.time(DF2 <- read.table("test.csv",header=TRUE,sep=",",quote="",
stringsAsFactors=FALSE,comment.char="",nrows=n,
colClasses=c("integer","integer","numeric",
"character","numeric","integer")))
# 1.2 sec (consistently). All known tricks and known nrows, see references.
system.time(DT <- fread("test.csv"))
# 0.1 sec (faster and friendlier)
identical(DF1, DF2)
all.equal(as.data.table(DF1), DT)
# Scaling up ...
l = vector("list",10)
for (i in 1:10) l[[i]] = DT
DTbig = rbindlist(l)
tables()
write.table(DTbig,"testbig.csv",sep=",",row.names=FALSE,quote=FALSE)
# 500MB csv (10 million rows x 6 columns)
system.time(DF <- read.table("testbig.csv",header=TRUE,sep=",",
quote="",stringsAsFactors=FALSE,comment.char="",nrows=1e7,
colClasses=c("integer","integer","numeric",
"character","numeric","integer")))
# 17.0 sec (varies)
system.time(DT <- fread("testbig.csv"))
# 0.8 sec
all(mapply(all.equal, DF, DT))
# Real data example (Airline data)
# https://stat-computing.org/dataexpo/2009/the-data.html
download.file("https://stat-computing.org/dataexpo/2009/2008.csv.bz2",
destfile="2008.csv.bz2")
# 109MB (compressed)
system("bunzip2 2008.csv.bz2")
# 658MB (7,009,728 rows x 29 columns)
colClasses = sapply(read.csv("2008.csv",nrows=100,stringsAsFactors=FALSE),class)
# 4 character, 24 integer, 1 logical. Incorrect.
colClasses = sapply(read.csv("2008.csv",nrows=200,stringsAsFactors=FALSE),class)
# 5 character, 24 integer. Correct. Might have missed data only using 100 rows
# since read.table assumes colClasses is correct.
system.time(DF <- read.table("2008.csv", header=TRUE, sep=",",
quote="",stringsAsFactors=FALSE,comment.char="",nrows=7009730,
colClasses=colClasses))
# 24.4 secs
system.time(DT <- fread("2008.csv"))
# 1.9 secs
table(sapply(DT,class))
# 5 character and 24 integer columns. Correct without needing to worry about colClasses
# issue above.
# Reads URLs directly :
fread("https://www.stats.ox.ac.uk/pub/datasets/csb/ch11b.dat")
# Decompresses .gz and .bz2 automatically :
fread("https://stat-computing.org/dataexpo/2009/1987.csv.bz2")
## End(Not run)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.