6 LECTURE: Getting and Cleaning Data
Optional Video
Datasets
Get the datasets for this chapter.
6.1 Reading and Writing Data
There are a few principal functions reading data into R.
read.table
,read.csv
, for reading tabular datareadLines
, for reading lines of a text filesource
, for reading in R code files (inverse
ofdump
)dget
, for reading in R code files (inverse
ofdput
)load
, for reading in saved workspacesunserialize
, for reading single R objects in binary form
There are of course, many R packages that have been developed to read in all kinds of other datasets, and you may need to resort to one of these packages if you are working in a specific area.
There are analogous functions for writing data to files
write.table
, for writing tabular data to text files (i.e. CSV) or connectionswriteLines
, for writing character data line-by-line to a file or connectiondump
, for dumping a textual representation of multiple R objectsdput
, for outputting a textual representation of an R objectsave
, for saving an arbitrary number of R objects in binary format (possibly compressed) to a file.serialize
, for converting an R object into a binary format for outputting to a connection (or file).
6.2 Reading Data Files with read.table()
The read.table()
function is one of the most commonly used functions
for reading data. The help file for read.table()
is worth reading in
its entirety if only because the function gets used a lot (run
?read.table
in R). I know, I know, everyone always says to read the
help file, but this one is actually worth reading.
The read.table()
function has a few important arguments:
file
, the name of a file, or a connectionheader
, logical indicating if the file has a header linesep
, a string indicating how the columns are separatedcolClasses
, a character vector indicating the class of each column in the datasetnrows
, the number of rows in the dataset. By defaultread.table()
reads an entire file.comment.char
, a character string indicating the comment character. This defalts to"#"
. If there are no commented lines in your file, it’s worth setting this to be the empty string""
.skip
, the number of lines to skip from the beginningstringsAsFactors
, should character variables be coded as factors? This defaults toTRUE
because back in the old days, if you had data that were stored as strings, it was because those strings represented levels of a categorical variable. Now we have lots of data that is text data and they don’t always represent categorical variables. So you may want to set this to beFALSE
in those cases. If you always want this to beFALSE
, you can set a global option viaoptions(stringsAsFactors = FALSE)
. I’ve never seen so much heat generated on discussion forums about an R function argument than thestringsAsFactors
argument. Seriously.
For small to moderately sized datasets, you can usually call read.table without specifying any other arguments
Note that foo.txt
is not a real dataset here. It is only used as an example for how to use read.table()
.
In this case, R will automatically
- skip lines that begin with a #
- figure out how many rows there are (and how much memory needs to be allocated)
- figure what type of variable is in each column of the table.
Telling R all these things directly makes R run faster and more
efficiently. The read.csv()
function is identical to read.table
except that some of the defaults are set differently (like the sep
argument).
6.3 Reading in Larger Datasets with read.table
With much larger datasets, there are a few things that you can do that will make your life easier and will prevent R from choking.
Read the help page for read.table, which contains many hints
Make a rough calculation of the memory required to store your dataset (see the next section for an example of how to do this). If the dataset is larger than the amount of RAM on your computer, you can probably stop right here.
Set
comment.char = ""
if there are no commented lines in your file.Use the
colClasses
argument. Specifying this option instead of using the default can make ’read.table’ run MUCH faster, often twice as fast. In order to use this option, you have to know the class of each column in your data frame. If all of the columns are “numeric”, for example, then you can just setcolClasses = "numeric"
. A quick an dirty way to figure out the classes of each column is the following:
initial <- read.table("datatable.txt", nrows = 100)
classes <- sapply(initial, class)
tabAll <- read.table("datatable.txt", colClasses = classes)
Note that datatable.txt
is not a real dataset here. It is only used as an example for how to use read.table()
.
- Set
nrows
. This doesn’t make R run faster but it helps with memory usage. A mild overestimate is okay. You can use the Unix toolwc
to calculate the number of lines in a file.
In general, when using R with larger datasets, it’s also useful to know a few things about your system.
- How much memory is available on your system?
- What other applications are in use? Can you close any of them?
- Are there other users logged into the same system?
- What operating system ar you using? Some operating systems can limit the amount of memory a single process can access
6.4 Calculating Memory Requirements for R Objects
Because R stores all of its objects physical memory, it is important to be cognizant of how much memory is being used up by all of the data objects residing in your workspace. One situation where it’s particularly important to understand memory requirements is when you are reading in a new dataset into R. Fortunately, it’s easy to make a back of the envelope calculation of how much memory will be required by a new dataset.
For example, suppose I have a data frame with 1,500,000 rows and 120 columns, all of which are numeric data. Roughly, how much memory is required to store this data frame? Well, on most modern computers double precision floating point numbers are stored using 64 bits of memory, or 8 bytes. Given that information, you can do the following calculation
1,500,000 × 120 × 8 bytes/numeric = 1,440,000,000 bytes
= 1,440,000,000 / 220 bytes/MB
= 1,373.29 MB
= 1.34 GB
So the dataset would require about 1.34 GB of RAM. Most computers these days have at least that much RAM. However, you need to be aware of
- what other programs might be running on your computer, using up RAM
- what other R objects might already be taking up RAM in your workspace
Reading in a large dataset for which you do not have enough RAM is one easy way to freeze up your computer (or at least your R session). This is usually an unpleasant experience that usually requires you to kill the R process, in the best case scenario, or reboot your computer, in the worst case. So make sure to do a rough calculation of memeory requirements before reading in a large dataset. You’ll thank me later.
6.5 Using the readr
Package
The readr
package is recently developed by RStudio to deal
with reading in large flat files quickly. The package provides
replacements for functions like read.table()
and read.csv()
. The
analogous functions in readr
are read_table()
and
read_csv()
. These functions are often much faster than their base
R analogues and provide a few other nice features such as progress
meters.
For the most part, you can read use read_table()
and read_csv()
pretty much anywhere you might use read.table()
and read.csv()
. In
addition, if there are non-fatal problems that occur while reading in
the data, you will get a warning and the returned data frame will have
some information about which rows/observations triggered the
warning. This can be very helpful for “debugging” problems with your
data before you get neck deep in data analysis.
The importance of the read_csv
function is perhaps better understood
from an historical perspective. R’s built in read.csv
function
similarly reads CSV files, but the read_csv
function in readr
builds on that by removing some of the quirks and “gotchas” of
read.csv
as well as dramatically optimizing the speed with which it
can read data into R. The read_csv
function also adds some nice
user-oriented features like a progress meter and a compact method for
specifying column types.
A typical call to read_csv
will look as follows.
##
## ── Column specification ────────────────────────────────────────────────────────
## cols(
## Standing = col_double(),
## Team = col_character()
## )
## # A tibble: 32 x 2
## Standing Team
## <dbl> <chr>
## 1 1 Spain
## 2 2 Netherlands
## 3 3 Germany
## 4 4 Uruguay
## 5 5 Argentina
## 6 6 Brazil
## 7 7 Ghana
## 8 8 Paraguay
## 9 9 Japan
## 10 10 Chile
## # … with 22 more rows
By default, read_csv
will open a CSV file and read it in line-by-line. It will also (by default), read in the first few rows of the table in order to figure out the type of each column (i.e. integer, character, etc.). From the read_csv
help page:
If ‘NULL’, all column types will be imputed from the first 1000 rows on the input. This is convenient (and fast), but not robust. If the imputation fails, you’ll need to supply the correct types yourself.
You can specify the type of each column with the col_types
argument.
In general, it’s a good idea to specify the column types explicitly. This rules out any possible guessing errors on the part of read_csv
. Also, specifying the column types explicitly provides a useful safety check in case anything about the dataset should change without you knowing about it.
Note that the col_types
argument accepts a compact representation. Here "cc"
indicates that the first column is character
and the second column is character
(there are only two columns). Using the col_types
argument is useful because often it is not easy to automatically figure out the type of a column by looking at a few rows (especially if a column has many missing values).
The read_csv
function will also read compressed files automatically. There is no need to decompress the file first or use the gzfile
connection function. The following call reads a gzip-compressed CSV file containing download logs from the RStudio CRAN mirror.
##
## ── Column specification ────────────────────────────────────────────────────────
## cols(
## date = col_date(format = ""),
## time = col_time(format = ""),
## size = col_double(),
## r_version = col_character(),
## r_arch = col_character(),
## r_os = col_character(),
## package = col_character(),
## version = col_character(),
## country = col_character(),
## ip_id = col_double()
## )
Note that the warnings indicate that read_csv
may have had some difficulty identifying the type of each column. This can be solved by using the col_types
argument.
## # A tibble: 10 x 10
## date time size r_version r_arch r_os package version country ip_id
## <chr> <chr> <int> <chr> <chr> <chr> <chr> <chr> <chr> <int>
## 1 2016-0… 22:00… 1.89e6 3.3.0 x86_64 mingw32 data.t… 1.9.6 US 1
## 2 2016-0… 22:00… 4.54e4 3.3.1 x86_64 mingw32 assert… 0.1 US 2
## 3 2016-0… 22:00… 1.43e7 3.3.1 x86_64 mingw32 stringi 1.1.1 DE 3
## 4 2016-0… 22:00… 1.89e6 3.3.1 x86_64 mingw32 data.t… 1.9.6 US 4
## 5 2016-0… 22:00… 3.90e5 3.3.1 x86_64 mingw32 foreach 1.4.3 US 4
## 6 2016-0… 22:00… 4.88e4 3.3.1 x86_64 linux-… tree 1.0-37 CO 5
## 7 2016-0… 22:00… 5.25e2 3.3.1 x86_64 darwin… surviv… 2.39-5 US 6
## 8 2016-0… 22:00… 3.23e6 3.3.1 x86_64 mingw32 Rcpp 0.12.5 US 2
## 9 2016-0… 22:00… 5.56e5 3.3.1 x86_64 mingw32 tibble 1.1 US 2
## 10 2016-0… 22:00… 1.52e5 3.3.1 x86_64 mingw32 magrit… 1.5 US 2
You can specify the column type in a more detailed fashion by using the various col_*
functions. For example, in the log data above, the first column is actually a date, so it might make more sense to read it in as a Date variable. If we wanted to just read in that first column, we could do
logdates <- read_csv("data/2016-07-19.csv.bz2",
col_types = cols_only(date = col_date()),
n_max = 10)
logdates
## # A tibble: 10 x 1
## date
## <date>
## 1 2016-07-19
## 2 2016-07-19
## 3 2016-07-19
## 4 2016-07-19
## 5 2016-07-19
## 6 2016-07-19
## 7 2016-07-19
## 8 2016-07-19
## 9 2016-07-19
## 10 2016-07-19
Now the date
column is stored as a Date
object which can be used for relevant date-related computations (for example, see the lubridate
package).
The read_csv
function has a progress
option that defaults to TRUE. This options provides a nice progress meter while the CSV file is being read. However, if you are using read_csv
in a function, or perhaps embedding it in a loop, it’s probably best to set progress = FALSE
.