Category Archives: R programming

Speed up your R code using a just-in-time (JIT) compiler

This post is about speeding up your R code using the JIT (just in time) compilation capabilities offered by the new (well, now a year old) {compiler} package. Specifically, dealing with the practical difference between enableJIT and the cmpfun functions.

If you do not want to read much, you can just skip to the example part.

As always, I welcome any comments to this post, and hope to update it when future JIT solutions will come along.

Continue reading

Do more with dates and times in R with lubridate 1.1.0

This is a guest post by Garrett Grolemund (mentored by Hadley Wickham)

Lubridate is an R package that makes it easier to work with dates and times. The newest release of lubridate (v 1.1.0) comes with even more tools and some significant changes over past versions. Below is a concise tour of some of the things lubridate can do for you. At the end of this post, I list some of the differences between lubridate (v 0.2.4) and lubridate (v 1.1.0). If you are an old hand at lubridate, please read this section to avoid surprises!

Lubridate was created by Garrett Grolemund and Hadley Wickham.

Parsing dates and times

Getting R to agree that your data contains the dates and times you think it does can be a bit tricky. Lubridate simplifies that. Identify the order in which the year, month, and day appears in your dates. Now arrange “y”, “m”, and “d” in the same order. This is the name of the function in lubridate that will parse your dates. For example,

library(lubridate)
ymd("20110604"); mdy("06-04-2011"); dmy("04/06/2011")
## "2011-06-04 UTC"
## "2011-06-04 UTC"
## "2011-06-04 UTC"

Parsing functions automatically handle a wide variety of formats and separators, which simplifies the parsing process.

If your date includes time information, add h, m, and/or s to the name of the function. ymd_hms() is probably the most common date time format. To read the dates in with a certain time zone, supply the official name of that time zone in the tz argument.

arrive < - ymd_hms("2011-06-04 12:00:00", tz = "Pacific/Auckland")
## "2011-06-04 12:00:00 NZST"
leave <- ymd_hms("2011-08-10 14:00:00", tz = "Pacific/Auckland")
## "2011-08-10 14:00:00 NZST"

Setting and Extracting information

Extract information from date times with the functions second(), minute(), hour(), day(), wday(), yday(), week(), month(), year(), and tz(). You can also use each of these to set (i.e, change) the given information. Notice that this will alter the date time. wday() and month() have an optional label argument, which replaces their numeric output with the name of the weekday or month.

second(arrive)
## 0
second(arrive) < - 25
arrive
## "2011-06-04 12:00:25 NZST"
second(arrive) <- 0
wday(arrive)
## 7
wday(arrive, label = TRUE)
## Sat

Time Zones

There are two very useful things to do with dates and time zones. First, display the same moment in a different time zone. Second, create a new moment by combining a given clock time with a new time zone. These are accomplished by with_tz() and force_tz().

For example, I spent last summer researching in Auckland, New Zealand. I arranged to meet with my advisor, Hadley, over skype at 9:00 in the morning Auckland time. What time was that for Hadley who was back in Houston, TX?

meeting < - ymd_hms("2011-07-01 09:00:00", tz = "Pacific/Auckland")
## "2011-07-01 09:00:00 NZST"
with_tz(meeting, "America/Chicago")
## "2011-06-30 16:00:00 CDT"

So the meetings occurred at 4:00 Hadley’s time (and the day before no less). Of course, this was the same actual moment of time as 9:00 in New Zealand. It just appears to be a different day due to the curvature of the Earth.

What if Hadley made a mistake and signed on at 9:00 his time? What time would it then be my time?

mistake < - force_tz(meeting, "America/Chicago")
## "2011-07-01 09:00:00 CDT"
with_tz(mistake, "Pacific/Auckland")
## "2011-07-02 02:00:00 NZST"

His call would arrive at 2:00 am my time! Luckily he never did that.

Continue reading

Printing nested tables in R – bridging between the {reshape} and {tables} packages

This post shows how to print a prettier nested pivot table, created using the {reshape} package (similar to what you would get with Microsoft Excel), so you could print it either in the R terminal or as a LaTeX table. This task is done by bridging between the cast_df object produced by the {reshape} package, and the tabular function introduced by the new {tables} package.

Here is an example of the type of output we wish to produce in the R terminal:

1
2
3
4
5
6
7
       ozone       solar.r        wind         temp       
 month mean  sd    mean    sd     mean   sd    mean  sd   
 5     23.62 22.22 181.3   115.08 11.623 3.531 65.55 6.855
 6     29.44 18.21 190.2    92.88 10.267 3.769 79.10 6.599
 7     59.12 31.64 216.5    80.57  8.942 3.036 83.90 4.316
 8     59.96 39.68 171.9    76.83  8.794 3.226 83.97 6.585
 9     31.45 24.14 167.4    79.12 10.180 3.461 76.90 8.356

Or in a latex document:

Motivation: creating pretty nested tables

In a recent post we learned how to use the {reshape} package (by Hadley Wickham) in order to aggregate and reshape data (in R) using the melt and cast functions.

The cast function is wonderful but it has one problem – the format of the output. As opposed to a pivot table in (for example) MS excel, the output of a nested table created by cast is very “flat”. That is, there is only one row for the header, and only one column for the row names. So for both the R terminal, or an Sweave document, when we deal with a more complex reshaping/aggregating, the result is not something you would be proud to send to a journal.

The opportunity: the {tables} package

The good news is that Duncan Murdoch have recently released a new package to CRAN called {tables}. The {tables} package can compute and display complex tables of summary statistics and turn them into nice looking tables in Sweave (LaTeX) documents. For using the full power of this package, you are invited to read through its detailed (and well written) 23 pages Vignette. However, some of us might have preferred to keep using the syntax of the {reshape} package, while also benefiting from the great formatting that is offered by the new {tables} package. For this purpose, I devised a function that bridges between cast_df (from {reshape}) and the tabular function (from {tables}).

The bridge: between the {tables} and the {reshape} packages

The code for the function is available on my github (link: tabular.cast_df.r on github) and it seems to works fine as far as I can see (though I wouldn’t run it on larger data files since it relies on melting a cast_df object.)

Here is an example for how to load and use the function:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
######################
# Loading the functions
######################
# Making sure we can source code from github
source("http://www.r-statistics.com/wp-content/uploads/2012/01/source_https.r.txt")
 
# Reading in the function for using tabular on a cast_df object:
source_https("https://raw.github.com/talgalili/R-code-snippets/master/tabular.cast_df.r")
 
 
 
######################
# example:
######################
 
############
# Loading and preparing some data
require(reshape)
names(airquality) <- tolower(names(airquality))
airquality2 <- airquality
airquality2$temp2 <- ifelse(airquality2$temp > median(airquality2$temp), "hot", "cold")
aqm <- melt(airquality2, id=c("month", "day","temp2"), na.rm=TRUE)
colnames(aqm)[4] <- "variable2"	# because otherwise the function is having problem when relying on the melt function of the cast object
head(aqm,3)
#  month day temp2 variable2 value
#1     5   1  cold     ozone    41
#2     5   2  cold     ozone    36
#3     5   3  cold     ozone    12
 
############
# Running the example:
tabular.cast_df(cast(aqm, month ~ variable2, c(mean,sd)))
tabular(cast(aqm, month ~ variable2, c(mean,sd))) # notice how we turned tabular to be an S3 method that can deal with a cast_df object
Hmisc::latex(tabular(cast(aqm, month ~ variable2, c(mean,sd)))) # this is what we would have used for an Sweave document

And here are the results in the terminal:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
> 
> tabular.cast_df(cast(aqm, month ~ variable2, c(mean,sd)))
 
       ozone       solar.r        wind         temp       
 month mean  sd    mean    sd     mean   sd    mean  sd   
 5     23.62 22.22 181.3   115.08 11.623 3.531 65.55 6.855
 6     29.44 18.21 190.2    92.88 10.267 3.769 79.10 6.599
 7     59.12 31.64 216.5    80.57  8.942 3.036 83.90 4.316
 8     59.96 39.68 171.9    76.83  8.794 3.226 83.97 6.585
 9     31.45 24.14 167.4    79.12 10.180 3.461 76.90 8.356
> tabular(cast(aqm, month ~ variable2, c(mean,sd))) # notice how we turned tabular to be an S3 method that can deal with a cast_df object
 
       ozone       solar.r        wind         temp       
 month mean  sd    mean    sd     mean   sd    mean  sd   
 5     23.62 22.22 181.3   115.08 11.623 3.531 65.55 6.855
 6     29.44 18.21 190.2    92.88 10.267 3.769 79.10 6.599
 7     59.12 31.64 216.5    80.57  8.942 3.036 83.90 4.316
 8     59.96 39.68 171.9    76.83  8.794 3.226 83.97 6.585
 9     31.45 24.14 167.4    79.12 10.180 3.461 76.90 8.356

And in an Sweave document:

Here is an example for the Rnw file that produces the above table:
cast_df to tabular.Rnw

I will finish with saying that the tabular function offers more flexibility then the one offered by the function I provided. If you find any bugs or have suggestions of improvement, you are invited to leave a comment here or inside the code on github.

(Link-tip goes to Tony Breyal for putting together a solution for sourcing r code from github.)

Merging two data.frame objects while preserving the rows’ order

Merging two data.frame objects in R is very easily done by using the merge function. While being very powerful, the merge function does not (as of yet) offer to return a merged data.frame that preserved the original order of, one of the two merged, data.frame objects.
In this post I describe this problem, and offer some easy to use code to solve it.

Let us start with a simple example:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
    x <- data.frame(
           ref = c( 'Ref1', 'Ref2' )
         , label = c( 'Label01', 'Label02' )
         )
    y <- data.frame(
          id = c( 'A1', 'C2', 'B3', 'D4' )
        , ref = c( 'Ref1', 'Ref2' , 'Ref3','Ref1' )
        , val = c( 1.11, 2.22, 3.33, 4.44 )
        )
 
#######################
# having a look at the two data.frame objects:
> x
   ref   label
1 Ref1 Label01
2 Ref2 Label02
> y
  id  ref  val
1 A1 Ref1 1.11
2 C2 Ref2 2.22
3 B3 Ref3 3.33
4 D4 Ref1 4.44

If we will now merge the two objects, we will find that the order of the rows is different then the original order of the “y” object. This is true whether we use “sort =T” or “sort=F”. You can notice that the original order was an ascending order of the “val” variable:

1
2
3
4
5
6
7
8
9
10
11
12
> merge( x, y, by='ref', all.y = T, sort= T)
   ref   label id  val
1 Ref1 Label01 A1 1.11
2 Ref1 Label01 D4 4.44
3 Ref2 Label02 C2 2.22
4 Ref3    <NA> B3 3.33
> merge( x, y, by='ref', all.y = T, sort=F )
   ref   label id  val
1 Ref1 Label01 A1 1.11
2 Ref1 Label01 D4 4.44
3 Ref2 Label02 C2 2.22
4 Ref3    <NA> B3 3.33

This is explained in the help page of ?merge:

The rows are by default lexicographically sorted on the common columns, but for ‘sort = FALSE’ are in an unspecified order.

Or put differently: sort=FALSE doesn’t preserve the order of any of the two entered data.frame objects (x or y); instead it gives us an
unspecified (potentially random) order.

However, it can so happen that we want to make sure the order of the resulting merged data.frame objects ARE ordered according to the order of one of the two original objects. In order to make sure of that, we could add an extra “id” (row index number) sequence on the dataframe we wish to sort on. Then, we can merge the two data.frame objects, sort by the sequence, and delete the sequence. (this was previously mentioned on the R-help mailing list by Bart Joosen).

Following is a function that implements this logic, followed by an example for its use:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
############## function:
	merge.with.order <- function(x,y, ..., sort = T, keep_order)
	{
		# this function works just like merge, only that it adds the option to return the merged data.frame ordered by x (1) or by y (2)
		add.id.column.to.data <- function(DATA)
		{
			data.frame(DATA, id... = seq_len(nrow(DATA)))
		}
		# add.id.column.to.data(data.frame(x = rnorm(5), x2 = rnorm(5)))
		order.by.id...and.remove.it <- function(DATA)
		{
			# gets in a data.frame with the "id..." column.  Orders by it and returns it
			if(!any(colnames(DATA)=="id...")) stop("The function order.by.id...and.remove.it only works with data.frame objects which includes the 'id...' order column")
 
			ss_r <- order(DATA$id...)
			ss_c <- colnames(DATA) != "id..."
			DATA[ss_r, ss_c]		
		}
 
		# tmp <- function(x) x==1; 1	# why we must check what to do if it is missing or not...
		# tmp()
 
		if(!missing(keep_order))
		{
			if(keep_order == 1) return(order.by.id...and.remove.it(merge(x=add.id.column.to.data(x),y=y,..., sort = FALSE)))
			if(keep_order == 2) return(order.by.id...and.remove.it(merge(x=x,y=add.id.column.to.data(y),..., sort = FALSE)))
			# if you didn't get "return" by now - issue a warning.
			warning("The function merge.with.order only accepts NULL/1/2 values for the keep_order variable")
		} else {return(merge(x=x,y=y,..., sort = sort))}
	}
 
######### example:
>     merge( x.labels, x.vals, by='ref', all.y = T, sort=F )
   ref   label id  val
1 Ref1 Label01 A1 1.11
2 Ref1 Label01 D4 4.44
3 Ref2 Label02 C2 2.22
4 Ref3    <NA> B3 3.33
>     merge.with.order( x.labels, x.vals, by='ref', all.y = T, sort=F ,keep_order = 1)
   ref   label id  val
1 Ref1 Label01 A1 1.11
2 Ref1 Label01 D4 4.44
3 Ref2 Label02 C2 2.22
4 Ref3    <NA> B3 3.33
>     merge.with.order( x.labels, x.vals, by='ref', all.y = T, sort=F ,keep_order = 2) # yay - works as we wanted it to...
   ref   label id  val
1 Ref1 Label01 A1 1.11
3 Ref2 Label02 C2 2.22
4 Ref3    <NA> B3 3.33
2 Ref1 Label01 D4 4.44

Here is a description for how to use the keep_order parameter:

keep_order can accept the numbers 1 or 2, in which case it will make sure the resulting merged data.frame will be ordered according to the original order of rows of the data.frame entered to x (if keep_order=1) or to y (if keep_order=2). If keep_order is missing, merge will continue working as usual. If keep_order gets some input other then 1 or 2, it will issue a warning that it doesn’t accept these values, but will continue working as merge normally would. Notice that the parameter “sort” is practically overridden when using keep_order (with the value 1 or 2).

The same code can be used to modify the original merge.data.frame function in base R, so to allow the use of the keep_order, here is a link to the patched merge.data.frame function (on github). If you can think of any ways to improve the function (or happen to notice a bug) please let me know either on github or in the comments. (also saying that you found the function to be useful will be fun to know about :) )

Update: Thanks to KY’s comment, I noticed the ?join function in the {plyr} library. This function is similar to merge (with less features, yet faster), and also automatically keeps the order of the x (first) data.frame used for merging, as explained in the ?join help page:

Unlike merge, (join) preserves the order of x no matter what join type is used. If needed, rows from y will be added to the bottom. Join is often faster than merge, although it is somewhat less featureful – it currently offers no way to rename output or merge on different variables in the x and y data frames.

reshaping data using melt and cast

Aggregation and Restructuring data (from “R in Action”)

The followings introductory post is intended for new users of R.  It deals with the restructuring of data: what it is and how to perform it using base R functions and the {reshape} package.

This is a guest article by Dr. Robert I. Kabacoff, the founder of (one of) the first online R tutorials websites: Quick-R. Kabacoff has recently published the book ”R in Action“, providing a detailed walk-through for the R language based on various examples for illustrating R’s features (data manipulation, statistical methods, graphics, and so on…). The previous guest post by Kabacoff introduced data.frame objects in R.

For readers of this blog, there is a 38% discount off the “R in Action” book (as well as all other eBooks, pBooks and MEAPs at Manning publishing house), simply by using the code rblogg38 when reaching checkout.

Let us now talk about the Aggregation and Restructuring of data in R:

Aggregation and Restructuring

R provides a number of powerful methods for aggregating and reshaping data. When you aggregate data, you replace groups of observations with summary statistics based on those observations. When you reshape data, you alter the structure (rows and columns) determining how the data is organized. This article describes a variety of methods for accomplishing these tasks.

We’ll use the mtcars data frame that’s included with the base installation of R. This dataset, extracted from Motor Trend magazine (1974), describes the design and performance characteristics (number of cylinders, displacement, horsepower, mpg, and so on) for 34 automobiles. To learn more about the dataset, see help(mtcars).

Transpose

The transpose (reversing rows and columns) is perhaps the simplest method of reshaping a dataset. Use the t() function to transpose a matrix or a data frame. In the latter case, row names become variable (column) names. An example is presented in the next listing.

Listing 1 Transposing a dataset

1
2
3
4
5
6
7
8
9
10
11
12
13
14
> cars <- mtcars[1:5,1:4]
> cars
                  mpg  cyl disp  hp
Mazda RX4         21.0   6  160 110
Mazda RX4 Wag     21.0   6  160 110
Datsun 710        22.8   4  108 93
Hornet 4 Drive    21.4   6  258 110
Hornet Sportabout 18.7   8  360 175
> t(cars)
     Mazda RX4 Mazda RX4 Wag Datsun 710 Hornet 4 Drive Hornet Sportabout
mpg         21        21           22.8           21.4              18.7
cyl          6         6            4.0            6.0               8.0
disp       160       160          108.0          258.0             360.0
hp         110       110           93.0           110.0            175.0

Listing 1 uses a subset of the mtcars dataset in order to conserve space on the page. You’ll see a more flexible way of transposing data when we look at the reshape package later in this article.

Aggregating data

It’s relatively easy to collapse data in R using one or more by variables and a defined function. The format is

1
aggregate(x, by, FUN)

where x is the data object to be collapsed, by is a list of variables that will be crossed to form the new observations, and FUN is the scalar function used to calculate summary statistics that will make up the new observation values.

As an example, we’ll aggregate the mtcars data by number of cylinders and gears, returning means on each of the numeric variables (see the next listing).

Listing 2 Aggregating data

1
2
3
4
5
6
7
8
9
10
11
12
13
> options(digits=3)
> attach(mtcars)
> aggdata <-aggregate(mtcars, by=list(cyl,gear), FUN=mean, na.rm=TRUE)
> aggdata
  Group.1 Group.2  mpg cyl disp  hp drat   wt qsec  vs   am gear carb
1       4       3 21.5   4  120  97 3.70 2.46 20.0 1.0 0.00    3 1.00
2       6       3 19.8   6  242 108 2.92 3.34 19.8 1.0 0.00    3 1.00
3       8       3 15.1   8  358 194 3.12 4.10 17.1 0.0 0.00    3 3.08
4       4       4 26.9   4  103  76 4.11 2.38 19.6 1.0 0.75    4 1.50
5       6       4 19.8   6  164 116 3.91 3.09 17.7 0.5 0.50    4 4.00
6       4       5 28.2   4  108 102 4.10 1.83 16.8 0.5 1.00    5 2.00
7       6       5 19.7   6  145 175 3.62 2.77 15.5 0.0 1.00    5 6.00
8       8       5 15.4   8  326 300 3.88 3.37 14.6 0.0 1.00    5 6.00

In these results, Group.1 represents the number of cylinders (4, 6, or 8) and Group.2 represents the number of gears (3, 4, or 5). For example, cars with 4 cylinders and 3 gears have a mean of 21.5 miles per gallon (mpg).

When you’re using the aggregate() function , the by variables must be in a list (even if there’s only one). You can declare a custom name for the groups from within the list, for instance, using by=list(Group.cyl=cyl, Group.gears=gear).

The function specified can be any built-in or user-provided function. This gives the aggregate command a great deal of power. But when it comes to power, nothing beats the reshape package.

The reshape package

The reshape package is a tremendously versatile approach to both restructuring and aggregating datasets. Because of this versatility, it can be a bit challenging to learn.

We’ll go through the process slowly and use a small dataset so that it’s clear what’s happening. Because reshape isn’t included in the standard installation of R, you’ll need to install it one time, using install.packages(“reshape”).

Basically, you’ll “melt” data so that each row is a unique ID-variable combination. Then you’ll “cast” the melted data into any shape you desire. During the cast, you can aggregate the data with any function you wish. The dataset you’ll be working with is shown in table 1.

Table 1 The original dataset (mydata)

ID

Time

X1

X2

1156
1235
2161
2224

 

In this dataset, the measurements are the values in the last two columns (5, 6, 3, 5, 6, 1, 2, and 4). Each measurement is uniquely identified by a combination of ID variables (in this case ID, Time, and whether the measurement is on X1 or X2). For example, the measured value 5 in the first row is uniquely identified by knowing that it’s from observation (ID) 1, at Time 1, and on variable X1.

Melting

When you melt a dataset, you restructure it into a format where each measured variable is in its own row, along with the ID variables needed to uniquely identify it. If you melt the data from table 1, using the following code

1
2
library(reshape)
md <- melt(mydata, id=(c("id", "time")))

You end up with the structure shown in table 2.

Table 2 The melted dataset

ID

Time

Variable

Value

11X15
12X13
21X16
22X12
11X26
12X25
21X21
22X24

 

Note that you must specify the variables needed to uniquely identify each measurement (ID and Time) and that the variable indicating the measurement variable names (X1 or X2) is created for you automatically.

Now that you have your data in a melted form, you can recast it into any shape, using the cast() function.

Casting

The cast() function starts with melted data and reshapes it using a formula that you provide and an (optional) function used to aggregate the data. The format is

1
newdata <- cast(md, formula, FUN)

Where md is the melted data, formula describes the desired end result, and FUN is the (optional) aggregating function. The formula takes the form

1
rowvar1 + rowvar2 + …  ~  colvar1 + colvar2 +

In this formula, rowvar1 + rowvar2 + … define the set of crossed variables that define the rows, and colvar1 + colvar2 + … define the set of crossed variables that define the columns. See the examples in figure 1. (click to enlarge the image)

Figure 1 Reshaping data with the melt() and cast() functions

Because the formulas on the right side (d, e, and f) don’t include a function, the data is reshaped. In contrast, the examples on the left side (a, b, and c) specify the mean as an aggregating function. Thus the data are not only reshaped but aggregated as well. For example, (a) gives the means on X1 and X2 averaged over time for each observation. Example (b) gives the mean scores of X1 and X2 at Time 1 and Time 2, averaged over observations. In (c) you have the mean score for each observation at Time 1 and Time 2, averaged over X1 and X2.

As you can see, the flexibility provided by the melt() and cast() functions is amazing. There are many times when you’ll have to reshape or aggregate your data prior to analysis. For example, you’ll typically need to place your data in what’s called long format resembling table 2 when analyzing repeated measures data (data where multiple measures are recorded for each observation).

Summary

Chapter 5 of R in Action reviews many of the dozens of mathematical, statistical, and probability functions that are useful for manipulating data. In this article, we have briefly explored several ways of aggregating and restructuring data.

 

This article first appeared as chapter 5.6 from the “R in action book, and is published with permission from Manning publishing house.  Other books in this serious which you might be interested in are (see the beginning of this post for a discount code):

data.frame objects in R (via “R in Action”)

The followings introductory post is intended for new users of R.  It deals with R data frames: what they are, and how to create, view, and update them.

This is a guest article by Dr. Robert I. Kabacoff, the founder of (one of) the first online R tutorials websites: Quick-R.  Kabacoff has recently published the book ”R in Action“, providing a detailed walk-through for the R language based on various examples for illustrating R’s features (data manipulation, statistical methods, graphics, and so on…)

For readers of this blog, there is a 38% discount off the “R in Action” book (as well as all other eBooks, pBooks and MEAPs at Manning publishing house), simply by using the code rblogg38 when reaching checkout.

Let us now talk about data frames:

Data Frames


A data frame is more general than a matrix in that different columns can contain different modes of data (numeric, character, and so on). It’s similar to the datasets you’d typically see in SAS, SPSS, and Stata. Data frames are the most common data structure you’ll deal with in R.

The patient dataset in table 1 consists of numeric and character data.

Table 1: A patient dataset

PatientID

AdmDate

Age

Diabetes

Status

110/15/200925Type1Poor
211/01/200934Type2Improved
310/21/200928Type1Excellent
410/28/200952Type1Poor

Because there are multiple modes of data, you can’t contain this data in a matrix. In this case, a data frame would be the structure of choice.

A data frame is created with the data.frame() function:

1
mydata <- data.frame(col1, col2, col3,…)

where col1, col2, col3, … are column vectors of any type (such as character, numeric, or logical). Names for each column can be provided with the names function.

The following listing makes this clear.

Listing 1 Creating a data frame

1
2
3
4
5
6
7
8
9
10
11
> patientID <- c(1, 2, 3, 4)
> age <- c(25, 34, 28, 52)
> diabetes <- c("Type1", "Type2", "Type1", "Type1")
> status <- c("Poor", "Improved", "Excellent", "Poor")
> patientdata <- data.frame(patientID, age, diabetes, status)
> patientdata
  patientID age diabetes status
1         1  25    Type1 Poor
2         2  34    Type2 Improved
3         3  28    Type1 Excellent
4         4  52    Type1 Poor

Each column must have only one mode, but you can put columns of different modes together to form the data frame. Because data frames are close to what analysts typically think of as datasets, we’ll use the terms columns and variables interchangeably when discussing data frames.

There are several ways to identify the elements of a data frame. You can use the subscript notation or you can specify column names. Using the patientdata data frame created earlier, the following listing demonstrates these approaches.

Listing 2 Specifying elements of a data frame

1
2
3
4
5
6
7
8
9
10
11
12
13
14
> patientdata[1:2]
  patientID age
1         1  25
2         2  34
3         3  28
4         4  52
> patientdata[c("diabetes", "status")]
  diabetes status
1    Type1 Poor
2    Type2 Improved
3    Type1 Excellent 
4    Type1 Poor
> patientdata$age    #age variable in the patient data frame
[1] 25 34 28 52

The $ notation in the third example is used to indicate a particular variable from a given data frame. For example, if you want to cross-tabulate diabetes type by status, you could use the following code:

1
2
3
4
5
> table(patientdata$diabetes, patientdata$status)
 
        Excellent Improved Poor
  Type1         1        0    2
  Type2         0        1    0

It can get tiresome typing patientdata$ at the beginning of every variable name, so shortcuts are available. You can use either the attach() and detach() or with() functions to simplify your code.

attach, detach, and with

The attach() function adds the data frame to the R search path. When a variable name is encountered, data frames in the search path are checked in order to locate the variable. Using a sample (mtcars) data frame, you could use the following code to obtain summary statistics for automobile mileage (mpg), and plot this variable against engine displacement (disp), and weight (wt):

1
2
3
summary(mtcars$mpg)
plot(mtcars$mpg, mtcars$disp)
plot(mtcars$mpg, mtcars$wt)

This could also be written as

1
2
3
4
5
attach(mtcars)
  summary(mpg)
  plot(mpg, disp)
  plot(mpg, wt)
detach(mtcars)

The detach() function removes the data frame from the search path. Note that detach() does nothing to the data frame itself. The statement is optional but is good programming practice and should be included routinely.

The limitations with this approach are evident when more than one object can have the same name. Consider the following code:

1
2
3
4
5
6
7
8
9
> mpg <- c(25, 36, 47)
> attach(mtcars)
 
The following object(s) are masked _by_ ‘.GlobalEnv: mpg
> plot(mpg, wt)
Error in xy.coords(x, y, xlabel, ylabel, log) :
  ‘x’ and ‘y’ lengths differ
> mpg
[1] 25 36 47

Here we already have an object named mpg in our environment when the mtcars data frame is attached. In such cases, the original object takes precedence, which isn’t what you want. The plot statement fails because mpg has 3 elements and disp has 32 elements. The attach() and detach() functions are best used when you’re analyzing a single data frame and you’re unlikely to have multiple objects with the same name. In any case, be vigilant for warnings that say that objects are being masked.

An alternative approach is to use the with() function. You could write the previous example as

1
2
3
4
5
with(mtcars, {
  summary(mpg, disp, wt)
  plot(mpg, disp)
  plot(mpg, wt)
})

In this case, the statements within the {} brackets are evaluated with reference to the mtcars data frame. You don’t have to worry about name conflicts here. If there’s only one statement (for example, summary(mpg)), the {} brackets are optional.

The limitation of the with() function is that assignments will only exist within the function brackets. Consider the following:

1
2
3
4
5
6
7
8
> with(mtcars, {
   stats <- summary(mpg)
   stats
  })
   Min. 1st Qu. Median Mean 3rd Qu. Max.
  10.40 15.43 19.20 20.09 22.80 33.90
> stats
Error: object ‘stats’ not found

If you need to create objects that will exist outside of the with() construct, use the special assignment operator <<- instead of the standard one (<-). It will save the object to the global environment outside of the with() call. This can be demonstrated with the following code:

1
2
3
4
5
6
7
8
9
> with(mtcars, {
   nokeepstats <- summary(mpg)
   keepstats <<- summary(mpg)
})
> nokeepstats
Error: object ‘nokeepstats’ not found
> keepstats
   Min. 1st Qu. Median Mean 3rd Qu. Max.
    10.40 15.43 19.20 20.09 22.80 33.90

Most books on R recommend using with() over attach(). I think that ultimately the choice is a matter of preference and should be based on what you’re trying to achieve and your understanding of the implications.

Case identifiers

In the patient data example, patientID is used to identify individuals in the dataset. In R, case identifiers can be specified with a rowname option in the data frame function. For example, the statement

1
2
patientdata <- data.frame(patientID, age, diabetes, status,
   row.names=patientID)

specifies patientID as the variable to use in labeling cases on various printouts and graphs produced by R.

Summary

One of the most challenging tasks in data analysis is data preparation. R provides various structures for holding data and many methods for importing data from both keyboard and external sources. One of those structures is data frames, which we covered here. Your ability to specify elements of these structures via the bracket notation is particularly important in selecting, subsetting, and transforming data.

R offers a wealth of functions for accessing external data. This includes data from flat files, web files, statistical packages, spreadsheets, and databases. Note that you can also export data from R into these external formats. We showed you how to use either the attach() and detach() or with() functions to simplify your code.

This article first appeared as chapter 2.2.4 from the “R in action book, and is published with permission from Manning publishing house.

Comparison of ave, ddply and data.table

A guest post by Paul Hiemstra.
————

Fortran and C programmers often say that interpreted languages like R are nice and all, but lack in terms of speed. How fast something works in R greatly depends on how it is implemented, i.e. which packages/functions does one use. A prime example, which shows up regularly on the R-help list, is letting a vector grow as you perform an analysis. In pseudo-code this might look like:

1
2
3
4
5
dum = NULL
for(i in 1:100000) {
   # new_outcome = ...do some stuff...
   dum = c(dum, new_outcome)
}

The problem here is that dum is continuously growing in size. This forces the operating system to allocate new memory space for the object, which is terribly slow. Preallocating dum to the length it is supposed to be greatly improves the performance. Alternatively, the use of apply type of functions, or functions from plyr package prevent these kinds of problems. But even between more advanced methods there are large differences between different implementations.

Take the next example. We create a dataset which has two columns, one column with values (e.g. amount of rainfall) and in the other a category (e.g. monitoring station id). We would like to know what the mean value is per category. One way is to use for loops, but I’ll skip that one for now. Three possibilities exist that I know of: ddply (plyr), ave (base R) and data.table. The piece of code at the end of this post compares these three methods. The outcome in terms of speed is:
(press the image to see a larger version)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
   datsize noClasses  tave tddply tdata.table
1    1e+05        10 0.091  0.035       0.011
2    1e+05        50 0.102  0.050       0.012
3    1e+05       100 0.105  0.065       0.012
4    1e+05       200 0.109  0.101       0.010
5    1e+05       500 0.113  0.248       0.012
6    1e+05      1000 0.123  0.438       0.012
7    1e+05      2500 0.146  0.956       0.013
8    1e+05     10000 0.251  3.525       0.020
9    1e+06        10 0.905  0.393       0.101
10   1e+06        50 1.003  0.473       0.100
11   1e+06       100 1.036  0.579       0.105
12   1e+06       200 1.052  0.826       0.106
13   1e+06       500 1.079  1.508       0.109
14   1e+06      1000 1.092  2.652       0.111
15   1e+06      2500 1.167  6.051       0.117
16   1e+06     10000 1.338 23.224       0.132

It is quite obvious that ddply performs very bad when the number of unique categories is large. The ave function performs better. However, the data.table option is by far the best one, outperforming both other alternatives easily. In response to this, Hadley Wickham (author of plyr) responded:

This is a drawback of the way that ddply always works with data frames. It will be a bit faster if you use summarise instead of data.frame (because data.frame is very slow), but I’m still thinking about how to overcome this fundamental limitation of the ddply approach.

I hope this comparison is of use to readers. And remember, think before complaining that R is slow :) .

Paul (e-mail: p.h.hiemstra@gmail.com)

ps This blogpost is based on discussions on the R-help and manipulatr mailing lists:
- http://www.mail-archive.com/r-help@r-project.org/msg142797.html
- http://groups.google.com/group/manipulatr/browse_thread/thread/5e8dfed85048df99

R code to perform the comparison

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
library(ggplot2)
library(data.table)
theme_set(theme_bw())
datsize = c(10e4, 10e5)
noClasses = c(10, 50, 100, 200, 500, 1000, 2500, 10e3)
comb = expand.grid(datsize = datsize, noClasses = noClasses)
res = ddply(comb, .(datsize, noClasses), function(x) {
  expdata = data.frame(value = runif(x$datsize),
                      cat = round(runif(x$datsize, min = 0, max = x$noClasses)))
  expdataDT = data.table(expdata)
 
  t1 = system.time(res1 <- with(expdata, ave(value, cat)))
  t2 = system.time(res2 <- ddply(expdata, .(cat), mean))
  t3 = system.time(res3 <- expdataDT[, sum(value), by = cat])
  return(data.frame(tave = t1[3], tddply = t2[3], tdata.table = t3[3]))
}, .progress = 'text')
 
res
 
ggplot(aes(x = noClasses, y = log(value), color = variable), data =
melt(res, id.vars = c("datsize","noClasses"))) + facet_wrap(~ datsize)
+ geom_line()

Managing a statistical analysis project – guidelines and best practices

In the past two years, a growing community of R users (and statisticians in general) have been participating in two major Question-and-Answer websites:

  1. The R tag page on Stackoverflow, and
  2. Stat over flow (which will soon move to a new domain, no worries, I’ll write about it once it happens)

In that time, several long (and fascinating) discussion threads where started, reflecting on tips and best practices for managing a statistical analysis project.  They are:

On the last thread in the list, the user chl, has started with trying to compile all the tips and suggestions together.  And with his permission, I am now republishing it here.  I encourage you to contribute from your own experience (either in the comments, or by answering to any of the threads I’ve linked to)

Continue reading