Chapter 5 Data Structures
One of R’s most powerful features is its ability to deal with tabular data -
such as you may already have in a spreadsheet or a CSV file. Let’s start by
looking at a toy dataset in your data/
directory, called feline-data.csv
:
The contents of the new file, feline-data.csv
:
5.1 The readr package
To read the data into R, we are going to use our first package, called readr. readr is part of a suite of packages called the “tidyverse” which were designed to work nicely together and to ease many common data operations.
The first time you use a package, you will need to install it (like installing an app on your phone from the app store). Additionally, it is a good idea to periodically check for updates to that package:
Everytime we want to use that package, you must load it into your R session, by
using the library
function:
Now we can load this data into R via the following using the read_csv
function,
and assign it to an object called cats
:
Parsed with column specification:
cols(
coat = col_character(),
weight = col_double(),
likes_string = col_double()
)
# A tibble: 3 x 3
coat weight likes_string
<chr> <dbl> <dbl>
1 calico 2.1 1
2 black 5 0
3 tabby 3.2 1
The read_csv
function is used for reading in tabular data stored in a text
file where the columns of data are separated by punctuation characters such as
CSV files (csv = comma-separated values). There is a base version of this called
read.csv
, but the readr
version (read_csv
) is a bit more user-friendly, and
uses more sensible defaults. Tabs and commas are the most common
punctuation characters used to separate or delimit data points in text files.
The object that is created by read_csv
is called a “data.frame” - a rectangular
table-like object with rows and columns.
We can begin exploring our dataset right away, first by looking at the whole thing:
# A tibble: 3 x 3
coat weight likes_string
<chr> <dbl> <dbl>
1 calico 2.1 1
2 black 5 0
3 tabby 3.2 1
And pulling out individual columns by specifying them using the $
operator:
[1] 2.1 5.0 3.2
[1] "calico" "black" "tabby"
We can do other operations on the columns:
[1] 4.1 7.0 5.2
[1] "My calico cat weighs 4.1 kilograms"
[2] "My black cat weighs 7 kilograms"
[3] "My tabby cat weighs 5.2 kilograms"
But what about
Error in cats$weight + cats$coat: non-numeric argument to binary operator
Understanding what happened here is key to successfully analyzing data in R.
5.2 Data Types
If you guessed that the last command will return an error because 2.1
plus
"black"
is nonsense, you’re right - and you already have some intuition for an
important concept in programming called data types. We can ask what type of
data something is:
[1] "double"
There are 4 main types:
double
/numeric
(decimal numbers),integer
(counting numbers),logical
(True/False),character
(free text)
[1] "double"
typeof(1L) # The L suffix forces the number to be an integer, since by default R uses double (decimal) numbers
[1] "integer"
[1] "logical"
[1] "character"
No matter how complicated our analyses become, all data in R is interpreted as one of these basic data types. This strictness has some really important consequences.
The table that R loaded our cats data into is something called a data.frame, and it is our first example of something called a data structure - that is, a structure which R knows how to build out of the basic data types.
We can see that it is a data.frame by calling the class
function on it:
[1] "spec_tbl_df" "tbl_df" "tbl" "data.frame"
5.3 Vectors and Type Coercion
To better understand this behavior, let’s meet another of the data structures: the vector.
If we are creating vectors on our own, we will normally use the c
(combine)
function:
[1] 1 3 5 7 9
A vector in R is essentially an ordered list of things, with the special condition that everything in the vector must be the same basic data type.
[1] "numeric"
This command indicates the basic data type found in this vector - in this case numeric
.
We can use the logical operators that we learned earlier with vectors:
[1] FALSE FALSE TRUE TRUE TRUE
Vectors can be any data type that we’ve already learned about. Let’s make a character vector:
[1] "Harry" "Ron" "Hermione" "Ginny"
[1] "character"
[1] TRUE FALSE FALSE FALSE
If we similarly do
[1] "numeric"
we see that cats$weight
is a vector, too - the columns of data we load into R
data.frames are all vectors, and that’s the root of why R forces everything in
a column to be the same basic data type.
5.3 Discussion 1
Why is R so opinionated about what we put in our columns of data? How does this help us?
5.3 Discussion 1
By keeping everything in a column the same, we allow ourselves to make simple assumptions about our data; if you can interpret one entry in the column as a number, then you can interpret all of them as numbers, so we don’t have to check every time. This consistency is what people mean when they talk about clean data; in the long run, strict consistency goes a long way to making our lives easier in R.
Given what we’ve learned so far, what do you think the following will produce?
This is something called type coercion, and it is the source of many surprises and the reason why we need to be aware of the basic data types and how R will interpret them. When R encounters a mix of types (here numeric and character) to be combined into a single vector, it will force them all to be the same type. Consider:
[1] "a" "TRUE"
[1] 0 1
The coercion rules go: logical
-> integer
-> double
/numeric
-> complex
->
character
, where -> can be read as are transformed into. You can try to
force coercion against this flow using the as.
functions:
[1] "0" "2" "4"
[1] 0 2 4
[1] FALSE TRUE TRUE
As you can see, some surprising things can happen when R forces one basic data type into another! Nitty-gritty of type coercion aside, the point is: if your data doesn’t look like what you thought it was going to look like, type coercion may well be to blame; make sure everything is the same type in your vectors and your columns of data.frames, or you will get nasty surprises!
But coercion can also be very useful! For example, in our cats
data
likes_string
is numeric, but we know that the 1s and 0s actually represent
TRUE
and FALSE
(a common way of representing them). We should use the
logical
datatype here, which has two states: TRUE
or FALSE
, which is
exactly what our data represents. We can ‘coerce’ this column to be logical
by
using the as.logical
function:
[1] 1 0 1
[1] TRUE FALSE TRUE
5.4 Data Frames
We said that columns in data.frames were vectors:
num [1:3] 2.1 5 3.2
logi [1:3] TRUE FALSE TRUE
These make sense. But what about
chr [1:3] "calico" "black" "tabby"
5.5 Factors
Another important data structure is called a factor. Factors usually look like character data, but are typically used to represent categorical information that have a defined set of values. For example, let’s make a vector of strings labelling cat colorations for all the cats in our study:
[1] "tabby" "tortoiseshell" "tortoiseshell" "black"
[5] "tabby"
We can turn a vector into a factor like so:
[1] "factor"
[1] tabby tortoiseshell tortoiseshell black tabby
Levels: black tabby tortoiseshell
Now R has noticed that there are three possible categories in our data - but it also did something surprising; instead of printing out the strings we gave it, we got a bunch of numbers instead. R has replaced our human-readable categories with numbered indices under the hood, this is necessary as many statistical calculations utilise such numerical representations for categorical data:
[1] "character"
[1] "integer"
5.5 Challenge 1
The default behaviour of
read.csv
is to convert all character columns to factors, which is one of the biggest reasons we preferread_csv
from thereadr
package. Look at the help for?read.csv
to figure out how to keep text columns as character vectors instead of factors; then write a command or two to show that the factor incats
is actually a character vector when loaded in this way.5.5 Solution to Challenge 1
One solution is use the argument
stringAsFactors
:cats <- read.csv(file="data/feline-data.csv") str(cats$coat) cats <- read.csv(file="data/feline-data.csv", stringsAsFactors=FALSE)
Another solution is use the argument
colClasses
that allow finer control.Note: new students find the help files difficult to understand; make sure to let them know that this is typical, and encourage them to take their best guess based on semantic meaning, even if they aren’t sure.
In modelling functions, it’s important to know what the baseline levels are. This is assumed to be the first factor, but by default factors are labelled in alphabetical order. You can change this by specifying the levels:
mydata <- c("case", "control", "control", "case")
factor_ordering_example <- factor(mydata, levels = c("control", "case"))
str(factor_ordering_example)
Factor w/ 2 levels "control","case": 2 1 1 2
In this case, we’ve explicitly told R that “control” should be represented by 1, and “case” by 2. This designation can be very important for interpreting the results of statistical models!
5.6 Lists
Another data structure you’ll want in your bag of tricks is the list
. A list
is simpler in some ways than the other types, because you can put anything you
want in it:
[[1]]
[1] 1
[[2]]
[1] "a"
[[3]]
[1] TRUE
[[4]]
[1] 1+4i
$title
[1] "Numbers"
$numbers
[1] 1 2 3 4 5 6 7 8 9 10
$data
[1] TRUE
We can now understand something a bit surprising in our data.frame; what happens if we run:
[1] "list"
We see that data.frames look like lists ‘under the hood’ - this is because a
data.frame is really a list of vectors and factors, as they have to be - in
order to hold those columns that are a mix of vectors and factors, the
data.frame needs something a bit more flexible than a vector to put all the
columns together into a familiar table. In other words, a data.frame
is a
special list in which all the vectors must have the same length.
In our cats
example, we have a character, a double and a logical variable. As
we have seen already, each column of data.frame is a vector.
[1] "calico" "black" "tabby"
# A tibble: 3 x 1
coat
<chr>
1 calico
2 black
3 tabby
[1] "list"
Classes 'tbl_df', 'tbl' and 'data.frame': 3 obs. of 1 variable:
$ coat: chr "calico" "black" "tabby"
Each row is an observation of different variables, itself a data.frame, and thus can be composed of elements of different types.
# A tibble: 1 x 3
coat weight likes_string
<chr> <dbl> <lgl>
1 calico 2.1 TRUE
[1] "list"
Classes 'tbl_df', 'tbl' and 'data.frame': 1 obs. of 3 variables:
$ coat : chr "calico"
$ weight : num 2.1
$ likes_string: logi TRUE
5.6 Challenge 3
There are several subtly different ways to call variables, observations and elements from data.frames:
cats[1]
cats[[1]]
cats$coat
cats["coat"]
cats[1, 1]
cats[, 1]
cats[1, ]
Try out these examples and explain what is returned by each one.
Hint: Use the function
typeof()
to examine what is returned in each case.5.6 Solution to Challenge 3
# A tibble: 3 x 1 coat <chr> 1 calico 2 black 3 tabby
We can think of a data frame as a list of vectors. The single brace
[1]
returns the first slice of the list, as another list. In this case it is the first column of the data frame.[1] "calico" "black" "tabby"
The double brace
[[1]]
returns the contents of the list item. In this case it is the contents of the first column, a vector of type factor.[1] "calico" "black" "tabby"
This example uses the
$
character to address items by name. coat is the first column of the data frame, again a vector of type factor.# A tibble: 3 x 1 coat <chr> 1 calico 2 black 3 tabby
Here we are using a single brace
["coat"]
replacing the index number with the column name. Like example 1, the returned object is a list.# A tibble: 1 x 1 coat <chr> 1 calico
This example uses a single brace, but this time we provide row and column coordinates. The returned object is the value in row 1, column 1. The object is an integer but because it is part of a vector of type factor, R displays the label “calico” associated with the integer value.
# A tibble: 3 x 1 coat <chr> 1 calico 2 black 3 tabby
Like the previous example we use single braces and provide row and column coordinates. The row coordinate is not specified, R interprets this missing value as all the elements in this column vector.
# A tibble: 1 x 3 coat weight likes_string <chr> <dbl> <lgl> 1 calico 2.1 TRUE
Again we use the single brace with row and column coordinates. The column coordinate is not specified. The return value is a list containing all the values in the first row.
5.7 Matrices
Last but not least is the matrix. We can declare a matrix full of zeros:
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 0 0 0 0 0 0
[2,] 0 0 0 0 0 0
[3,] 0 0 0 0 0 0
And similar to other data structures, we can ask things about our matrix:
[1] "matrix"
[1] "double"
num [1:3, 1:6] 0 0 0 0 0 0 0 0 0 0 ...
[1] 3 6
[1] 3
[1] 6
5.7 Challenge 4
What do you think will be the result of
length(matrix_example)
? Try it. Were you right? Why / why not?5.7 Solution to Challenge 4
What do you think will be the result of
length(matrix_example)
?[1] 18
Because a matrix is a vector with added dimension attributes,
length
gives you the total number of elements in the matrix.
5.7 Challenge 5
Make another matrix, this time containing the numbers 1:50, with 5 columns and 10 rows. Did the
matrix
function fill your matrix by column, or by row, as its default behaviour? See if you can figure out how to change this. (hint: read the documentation formatrix
!)5.7 Solution to Challenge 5
Make another matrix, this time containing the numbers 1:50, with 5 columns and 10 rows. Did the
matrix
function fill your matrix by column, or by row, as its default behaviour? See if you can figure out how to change this. (hint: read the documentation formatrix
!)
5.7 Challenge 6
Create a list of length two containing a character vector for each of the sections in this part of the workshop:
- Data types
- Data structures
Populate each character vector with the names of the data types and data structures we’ve seen so far.
5.7 Solution to Challenge 6
dataTypes <- c('double', 'complex', 'integer', 'character', 'logical') dataStructures <- c('data.frame', 'vector', 'factor', 'list', 'matrix') answer <- list(dataTypes, dataStructures)
Note: it’s nice to make a list in big writing on the board or taped to the wall listing all of these types and structures - leave it up for the rest of the workshop to remind people of the importance of these basics.
5.7 Challenge 7
Consider the R output of the matrix below:
[,1] [,2] [1,] 4 1 [2,] 9 5 [3,] 10 7
What was the correct command used to write this matrix? Examine each command and try to figure out the correct one before typing them. Think about what matrices the other commands will produce.
matrix(c(4, 1, 9, 5, 10, 7), nrow = 3)
matrix(c(4, 9, 10, 1, 5, 7), ncol = 2, byrow = TRUE)
matrix(c(4, 9, 10, 1, 5, 7), nrow = 2)
matrix(c(4, 1, 9, 5, 10, 7), ncol = 2, byrow = TRUE)
5.7 Solution to Challenge 7
Consider the R output of the matrix below:
[,1] [,2] [1,] 4 1 [2,] 9 5 [3,] 10 7
What was the correct command used to write this matrix? Examine each command and try to figure out the correct one before typing them. Think about what matrices the other commands will produce.