Content
How do I create a libname in R?
Q: I have a directory full of datasets. I need to use several of them in my analysis. In SAS®, I would create a libname so I could access all of them. Is there a way to do something similar in R?
A: With the libr package, you can create a libname in R very much the same way you create a libname in SAS®.
The above statement will create a libname “mylib” from the directory specified on the second parameter. The libname will use the CSV engine. If there are any CSV files in the directory, they will be all loaded into the library. To work directly with the datasets, you can then do:
To access your datasets.
Which data formats does the libname function support?
Q: I can see from the examples that the libr package supports CSV and SAS dataset file formats. What other data formats does the package support?
A: The package supports the following data formats:
csv, sas7bdat, rds, Rdata, rda, xls, xlsx, xpt, and dbf. The
libname() help page has a full list, and a short discussion
of some details on each format. Note that the sas7bdat file format is
read-only at this time.
Is there a way to filter the datasets in my libname?
Q: I have a directory with over 100 datasets. I want
to use the libname() function, but worry about loading all
those datasets into memory. Is there a way I can filter the libname, to
get only some of the datasets?
A: Yes. The filter parameter on the
libname() function allows you to pass a wildcard filter
string. For example, the following call will load only those datasets
that start with ‘a’:
If you have a more complicated filter criteria, you can also pass a vector of filter strings. The below example will load only those datasets that start with ‘a’ or ‘b’.
How do I view the variables in my datasets?
Q: I’m doing some analysis with my data, and can’t remember all the variable names. Is there an easy way to view or print out the variables in my datasets?
A: Yes. The dictionary() function from
the libr pacakge will return a dataset with all the
variables in your dataset, and some interesting attributes for each
variable. The dictionary() function works on a single data
frame, or an entire library. You can save this dictionary as metadata,
print it, or even create a report from it. Here is an example:
# Create libname
libname(mylib, "c:/mypath/mydata", "csv")
# Get dictionary
d <- dictionary(mylib)
# View dictionary
# View(d)How do I export data to another file format?
Q: Let’s say I have some data in one format (sas7bdat), and want to export this data to another format (csv or Excel). How can I do that with the libr package?
A: The lib_export() function was
designed for this purpose. You can take an existing library and export
the entire thing to another library with a different file format. Like
this:
The above statements will take the SAS® datasets in the library “libA”, export them to CSV, place the new CSV files in the directory “c:/mypath/mydata2”, and assign a new libname “libB” to that directory. You now have two libnames, and can continue working with each as desired.
How do I copy a library?
Q: I have a directory full of datasets. I want to back up the entire thing to another directory. How can I do that?
A: You can use the lib_copy() function,
like this:
# Create libname
libname(lib1, "c:/mypath/mydata1", "csv")
# Copy to a new location
lib_copy(lib1, lib2, "c:/mypath/mydata2")You will now have a reference to the new libname lib2 at
the new location, and can use this libname like any other.
Can I really do a datastep in R?
Q: When I first started learning R I searched all over for a way to do a datastep. I was shocked to learn there was nothing similar. Does the libr package really allow me to do a datastep in R?
A: Yes. The libr
datastep() function does not have all the capabilities of a
SAS® datastep. But it has the most commonly-used functionality. You can
loop through the data row by row, examine, and compare variable values
for each row. It has basic data shaping, grouping, retain, assigning of
attributes, and a datastep array. Here is a simple example showing
categorization of an age variable into age groups:
library(dplyr)
library(libr)
# Define data library
libname(dat, "./data", "csv")
# Prepare data
dm_mod <- dat$DM %>%
select(USUBJID, SEX, AGE, ARM) %>%
filter(ARM != "SCREEN FAILURE") %>%
datastep({
if (AGE >= 18 & AGE <= 24)
AGECAT = "18 to 24"
else if (AGE >= 25 & AGE <= 44)
AGECAT = "25 to 44"
else if (AGE >= 45 & AGE <= 64)
AGECAT <- "45 to 64"
else if (AGE >= 65)
AGECAT <- ">= 65"
}) The datastep example above is part of a dplyr pipeline, but it can also function independently. Notice that, just like a SAS® datastep, you don’t have to declare new variables. You can just assign the new variable a value, and the datastep function will create it automatically.
You can check out the datastep() help page, or the datastep
vignette for additional examples and complete documentation.
Why is the datastep so slow?
Q: I like the datastep() function very
much. But it seems quite slow. Is there anything I can do to speed it
up?
A: Yes. Performance of the datastep()
is directly related to the size of the input data. The best thing you
can do to increase performance is to reduce the input data to only those
rows and columns that you need. The Base R subset()
function and Tidyverse select() and
filter() functions are useful for this purpose. Or you can
use the Base R subset brackets ([]) if you are familiar with that
syntax. If the datastep performance is still not satisfactory, it is
recommended that you explore other R functions to perform your intended
operation.
Can I do “set” and “merge” operations with the datastep?
Q: In SAS®, I used the datastep frequently to combine two or more datasets. Does the libr datastep support “set” and “merge”?
A: Yes. The datastep() function
supports both “set” and “merge” operations. The “set” parameter accepts
a list of one or more datasets to stack together, and the “merge”
parameters are used in almost the same way as SAS®. Here is an
example:
# Subset iris dataset
dat1 <- subset(mtcars, cyl == 4, c('mpg', 'cyl', 'disp'))[1:5, ]
dat2 <- subset(mtcars, cyl == 6, c('mpg', 'cyl', 'disp'))[1:5, ]
dat3 <- mtcars[1:10, c('hp', 'drat', 'wt')]
# Stack datasets using set operation
res1 <- datastep(dat1, set = dat2, {})
# mpg cyl disp
# 1 22.8 4 108.0
# 2 24.4 4 146.7
# 3 22.8 4 140.8
# 4 32.4 4 78.7
# 5 30.4 4 75.7
# 6 21.0 6 160.0
# 7 21.0 6 160.0
# 8 21.4 6 258.0
# 9 18.1 6 225.0
# 10 19.2 6 167.6
# Merge row by row
res2 <- datastep(res1, merge = dat3, {})
# mpg cyl disp hp drat wt
# 1 22.8 4 108.0 110 3.90 2.620
# 2 24.4 4 146.7 110 3.90 2.875
# 3 22.8 4 140.8 93 3.85 2.320
# 4 32.4 4 78.7 110 3.08 3.215
# 5 30.4 4 75.7 175 3.15 3.440
# 6 21.0 6 160.0 105 2.76 3.460
# 7 21.0 6 160.0 245 3.21 3.570
# 8 21.4 6 258.0 62 3.69 3.190
# 9 18.1 6 225.0 95 3.92 3.150
# 10 19.2 6 167.6 123 3.92 3.440The above merge shows how you can append columns even without a key
column. If you want to merge by a key, use the “merge_by” and “merge_in”
parameters. See the datastep() documentation for more
information and examples.