05-data-structures-part2.Rmd

---
layout: page
title: R for reproducible scientific analysis
subtitle: Data frames and reading in data
minutes: 45
---

```{r, include=FALSE}
source("tools/chunk-options.R")
# Silently load in the data so the rest of the lesson works
library(gapminder)
#gapminder <- read.csv("data/gapminder-FiveYearData.csv", header=TRUE)
```

> ## Learning Objectives {.objectives}
>
> * Become familiar with data frames
> * Be able to read regular data into R
>

## Data frames

Data frames are similar to matrices, except each column can be a
different atomic type. A data frames is the standard structure for
storing and manipulating rectangular data sets.  Underneath the hood,
data frames are really lists, where each element is an atomic vector,
with the added restriction that they're all the same length.  As you
will see, if we pull out one column of a data frame,we will have a
vector. You will probably find that data frames are more complicated
than vectors and other data structures we have considered, but they
provide powerful capabilities.


Data frames can be created manually with the `data.frame` function:

```{r}
df <- data.frame(id = c('a', 'b', 'c', 'd', 'e', 'f'), x = 1:6, y = c(214:219))
df
```

> ## Challenge 1: Data frames {.challenge}
>
> Try using the `length` function to query
> your data frame `df`. Does it give the result
> you expect?
>

Each column in the data frame is simply a list element, which is why
when you ask for the `length` of the data frame, it tells you the
number of columns. If you actually want the number of rows, you can
use the `nrow` function.

We can add columns or rows to a data.frame using `cbind` or `rbind`
(these are the two-dimensional equivalents of the `c` function):

## Adding columns to a data frame

To add a column we can use `cbind`:

```{r}
df <- cbind(df, 6:1)
df
```

Note that R automatically names the column. We may decide to change
the name by assigning a value using the `names` function:

```{r}
names(df)[4] <- 'SixToOne'
```

We can also provide a name when we add the column:

```{r}
df <- cbind(df, caps=LETTERS[1:6])
df
```

(`LETTERS` and `letters` are built-in constants.)


## Adding rows to a data frame 

To add a row we use `rbind`:

```{r}
df <- rbind(df, list("g", 11, 42, 0, "G"))
```

Note that we add the row as a list, because we have multiple types
across the columns. Nevertheless, this doesn't work as expected! What
do the error messages tell us?

It appears that R was trying to append "g" and "G" as factor
levels. Why? Let's examine the first column.  We can access a column
in a `data.frame` by using the `$` operator.


```{r}
class(df$id)
str(df)
```

When we created the data frame, R automatically made the first
and last columns into factors, not character vectors. There are no
pre-existing levels "g" and "G", so the attempt to add these values
fails. A row was added to the data frame, only there are missing
values in the factor columns:

```{r}
df
```


There are two ways we can work around this issue:

1. We can convert the factor columns into characters. This is
convenient, but we lose the factor structure. 

2. We can add factor levels to accomodate the new additions. If we
really do want the columns to be factors, this is the correct way to
proceed.

We will illustrate both solutions in the same data frame:

```{r}
df$id <- as.character(df$id)  # convert to character
class(df$id)
levels(df$caps) <- c(levels(df$caps), 'G') # add a factor level
class(df$caps)
```

Okay, now let's try adding that row again.

```{r}
df <- rbind(df, list("g", 11, 42, 0, 'G'))
tail(df, n=3)
```

We successfully added the last row, but we have an undesired row with
two `NA` values. How do delete it?

## Deleting rows and handling NA


There are multiple ways to delete a row containing `NA`:

```{r}
df[-7, ]  # The minus sign tells R to delete the row
df[complete.cases(df), ]  # Return `TRUE` when no missing values
na.omit(df)  # Another function for the same purpose
df <- na.omit(df)
```

## Combining data frames

We  can also row-bind data.frames together, but notice what happens to
the rownames:

```{r}
rbind(df, df)
```

R is making sure that rownames are unique. You can restore sequential
numbering by setting rownames to NULL:

```{r}
df2 <- rbind(df, df)
rownames(df2) <- NULL
df2
```


<!-- 
When we add a row we need to use a list, because each column is
a different type!  If you want to add multiple rows to a data.frame,
you will need to separate the new columns in the list:

```{r}
df <- rbind(df, list(c("l", "m"), c(12, 13), c(534, -20), c(-1, -2),
c('H', 'I')))
tail(df, n=3)
```
-->

> ## Challenge 2 {.challenge}
>
> Create a data frame that holds the following information for yourself:
>
> * First name
> * Last name
> * Age
>
> Then use rbind to add the same information for the people sitting near you.
>
> Now use cbind to add a column of logicals that will, for each
>person, hold the answer to the question,
> "Is there anything in this workshop you're finding confusing?"
>

## Reading in data

Remember earlier we obtained the gapminder dataset.
If you're curious about where this data comes from you might like to
look at the [Gapminder website](http://www.gapminder.org/data/documentation/).

Now we want to load the gapminder data into R.
Before reading in data, it's a good idea to have a look at its structure.
Let's take a look using our newly-learned shell skills:

```{r, engine='bash'}
cd data/ # navigate to the data directory of the project folder
head gapminder-FiveYearData.csv
```

As its file extension would suggest, the file contains comma-separated
values, and seems to contain a header row.

We can use `read.table` to read this into R

```{r}
gapminder <- read.table(
  file="data/gapminder-FiveYearData.csv",
  header=TRUE, sep=","
)
head(gapminder)
```

Because we know the structure of the data, we're able
to specify the appropriate arguments to `read.table`.
Without these arguments, `read.table` will do its
best to do something sensible, but it is always more
reliable to explicitly tell `read.table` the structure
of the data. `read.csv` function provides a convenient shortcut
for loading in CSV files.

```{r}
# Here is the read.csv version of the above command
gapminder <- read.csv(
  file="data/gapminder-FiveYearData.csv"
)
head(gapminder)
```

> ## Miscellaneous Tips {.callout}
>
> 1. Another type of file you might encounter are tab-separated
> format. To specify a tab as a separator, use `"\t"`.
>
> 2. You can also read in files from the Internet by replacing
> the file paths with a web address.
>
> 3. You can read directly from excel spreadsheets without
> converting them to plain text first by using the `xlsx` package.
>

To make sure our analysis is reproducible, we should put the code
into a script file so we can come back to it later.

> ## Challenge 3 {.challenge}
>
> Go to file -> new file -> R script, and write an R script
> to load in the gapminder dataset. Put it in the `scripts/`
> directory and add it to version control.
>
> Run the script using the `source` function, using the file path
> as its argument (or by pressing the "source" button in RStudio).
>

## Using data frames: the `gapminder` dataset


To recap what we've just learned, let's have a look at our example
data (life expectancy in various countries for various years).

Remember, there are a few functions we can use to interrogate data structures in R:

```{r, eval=FALSE}
class() # what is the data structure?
typeof() # what is its atomic type?
length() # how long is it? What about two dimensional objects?
attributes() # does it have any metadata?
str() # A full summary of the entire object
dim() # Dimensions of the object - also try nrow(), ncol()
```

Let's use them to explore the gapminder dataset.

```{r}
typeof(gapminder)
```

Remember, data frames are lists 'under the hood'.

```{r}
class(gapminder)
```

The gapminder data is stored in a "data.frame". This is the default
data structure when you read in data, and (as we've heard) is useful
for storing data with mixed types of columns.

Let's look at some of the columns.

> ## Challenge 4: Data types in a real dataset {.challenge}
>
> Look at the first 6 rows of the gapminder data frame we loaded before:
>
> ```{r}
> head(gapminder)
> ```
>
> Write down what data type you think is in each column
>


```{r}
typeof(gapminder$year)
typeof(gapminder$lifeExp)
```

Can anyone guess what we should expect the type of the continent column to be?

```{r}
typeof(gapminder$continent)
```

If you were expecting a the answer to be "character", you would rightly be
surprised by the answer. Let's take a look at the class of this column:

```{r}
class(gapminder$continent)
```

One of the default behaviours of R is to treat any text columns as "factors"
when reading in data. The reason for this is that text columns often represent
categorical data, which need to be factors to be handled appropriately by
the statistical modeling functions in R.

However it's not obvious behaviour, and something that trips many people up. We can
disable this behaviour and read in the data again.

```{r}
options(stringsAsFactors=FALSE)
gapminder <- read.table(
  file="data/gapminder-FiveYearData.csv", header=TRUE, sep=","
)
```

Remember, if you do turn it off automatic conversion to factors you will need to
explicitly convert the variables into factors when
running statistical models.
This can be useful, because it forces you to think about the
question you're asking, and makes it easier to specify the ordering of the categories.

However there are many in the R community who find it more sensible to
leave this as the default behaviour.

> ## Tip: Changing options {.callout} 
> When R starts, it runs any
>code in the file `.Rprofile` in the project directory. You can make
>permanent changes to default behaviour by storing the changes in that
>file. BE CAREFUL, however. If you change R's default options,
>programs written by others may not run correctly in your environment
>and vice versa. For this reason, some argue that most such changes
>should be in your scripts, where they are visible.

The first thing you should do when reading data in, is check that it matches what
you expect, even if the command ran without warnings or errors. The `str` function,
short for "structure", is really useful for this:

```{r}
str(gapminder)
```

We can see that the object is a `data.frame` with 1,704 observations (rows),
and 6 variables (columns). Below that, we see the name of each column, followed
by a ":", followed by the type of variable in that column, along with the first
few entries.

As discussed above, we can retrieve or modify the column or row names
of the data.frame:

```{r}
colnames(gapminder)  
copy <- gapminder
colnames(copy) <- letters[1:6]
head(copy, n=3)
```

> ## Challenge 5 {.challenge}
>
> Recall that we also used the `names` function (above) to modify
> column names. Does it matter which you use? You can check help with
> `?names` and `?colnames` to see whether it should matter.


```{r}
rownames(gapminder)[1:20]
```

See those numbers in the square brackets on the left? That tells you
the number of the first entry in that row of output. So we see that
for the 5th row, the rowname is "5". In this case, the rownames are
simply the row numbers.


There are a few related ways of retrieving and modifying this information.
`attributes` will give you both the row and column names, along with the
class information, while `dimnames` will give you just the rownames and
column names.

In both cases, the output object is stored in a `list`:

```{r}
str(dimnames(gapminder))
```

## Understanding how lists are used in function output


Lets run a basic linear regression on the gapminder dataset:

```{r}
# What is the relationship between life expectancy and year?
reg <- lm(lifeExp ~ year, data=gapminder)
```

We won't go into too much detail, but briefly: 

* `lm` estimates linear statistical models
* The first argument is a formula, with  `a ~ b` meaning that `a`,
     the dependent (or response) variable, is a
    function of `b`, the independent variable. 
* We tell `lm` to use the gapminder data frame, so it knows where to
 find the variables `lifeExp` and `year`. 



Let's look at the output:

```{r}
reg
```

Not much there right? But if we look at the structure...

```{r}
str(reg)
```

There's a great deal stored in nested lists! The structure function
allows you to see all the data available, in this case, the data that
was returned by the `lm` function.

For now, we can look at the `summary`:

```{r}
summary(reg)
```

As you might expect, life expectancy has slowly been increasing over
time, so we see a significant positive association!


## Challenge Solutions

> ## Solution to Challenge 2 {.challenge}
>
> Create a data frame that holds the following information for yourself:
>
> * First name
> * Last name
> * Age
>
> Then use rbind to add the same information for the people sitting near you.
>
> Now use cbind to add a column of logicals answering the question,
> "Is there anything in this workshop you're finding confusing?"
>
> ```{r, eval=FALSE}
> my_df <- data.frame(first_name = "Software", last_name = "Carpentry", age = 17)
> my_df <- rbind(my_df, list("Jane", "Smith", 29))
> my_df <- rbind(my_df, list(c("Jo", "John"), c("White", "Lee"), c(23, 41)))
> my_df <- cbind(my_df, confused = c(FALSE, FALSE, TRUE, FALSE))
> ```

> ## Solution to Challenge 5 {.challenge}
> 
> `?colnames` tells you that the `colnames` function is the same as
> `names` for a data frame. For other structures, they may not be the
> same. In particular, `names` does not work for matrices, but
> `colnames` does. You can verify this with 
> ```{r}
> m <- matrix(1:9, nrow=3)
> colnames(m) <- letters[1:3] # works as you would expect
> names(m) <- letters[1:3]  # destroys the matrix
> ```