r4ds/rectangle.qmd

# Data rectangling {#sec-rectangle-data}

```{r}
#| results: "asis"
#| echo: false
source("_common.R")
status("drafting")
```

## Introduction

In this chapter, you'll learn the art of data **rectangling**, taking data that is fundamentally tree-like and converting it into a rectangular data frames made up of rows and columns.
This is important because hierarchical data is surprisingly common, especially when working with data that comes from a web API.

To learn about rectangling, you'll first learn about lists, the data structure that makes hierarchical data possible in R.
Then you'll learn about two crucial tidyr functions: `tidyr::unnest_longer()`, which converts children in rows, and `tidyr::unnest_wider()`, which converts children into columns.
We'll then show you a few case studies, applying these simple function multiple times to solve real complex problems.
We'll finish off by talking about JSON, the most frequent source of hierarchical datasets and common format for data exchange on the web.

### Prerequisites

In this chapter we'll continue using tidyr.
We'll also use repurrrsive to supply some interesting datasets to practice your rectangling skills, and we'll finish up with a little jsonlite, which we'll use to read JSON files into R lists.

```{r}
#| label: setup
#| message: false

library(tidyverse)
library(repurrrsive)
library(jsonlite)
```

## Lists

So far we've used simple vectors like integers, numbers, characters, date-times, and factors.
These vectors are simple because they're homogeneous: every element is same type.
If you want to store element of different types, you need a **list**, which you create with `list()`:

```{r}
x1 <- list(1:4, "a", TRUE)
x1
```

It's often convenient to name the components, or **children**, of a list, which you can do in the same way as naming the columns of a tibble:

```{r}
x2 <- list(a = 1:2, b = 1:3, c = 1:4)
x2
```

Even for these very simple lists, printing takes up quite a lot of space.
A useful alternative is `str()`, which generates a compact display of the **str**ucture, de-emphasizing the contents:

```{r}
str(x1)
str(x2)
```

As you can see, `str()` displays each child on its own line.
It displays the name, if present, then an abbreviation of the type, then the first few values.

### Hierarchy

Lists can contain any type of object, including other lists.
This makes them suitable for representing hierarchical or tree-like structures:

```{r}
x3 <- list(list(1, 2), list(3, 4))
str(x3)
```

This is notably different to `c()`, which generates a flat vector:

```{r}
c(c(1, 2), c(3, 4))

x4 <- c(list(1, 2), list(3, 4))
str(x4)
```

As lists get more complex, `str()` gets more useful, as it lets you see the hierarchy at a glance:

```{r}
x5 <- list(1, list(2, list(3, list(4, list(5)))))
str(x5)
```

As lists get even large and more complex, even `str()` starts to fail, you'll need to switch to `View()`[^rectangle-1].
@fig-view-collapsed shows the result of calling `View(x4)`. The viewer starts by showing just the top level of the list, but you can interactively expand any of the components to see more, as in @fig-view-expand-1. RStudio will also show you the code you need to access that element, as in @fig-view-expand-2. We'll come back to how this code works in @sec-vector-subsetting.

[^rectangle-1]: This is an RStudio feature.

```{r}
#| label: fig-view-collapsed
#| fig.cap: >
#|   The RStudio allows you to interactively explore a complex list.  
#|   The viewer opens showing only the top level of the list.
#| echo: false
#| out-width: NULL
knitr::include_graphics("screenshots/View-1.png", dpi = 220)
```

```{r}
#| label: fig-view-expand-1
#| fig.cap: >
#|   Clicking on the rightward facing triangle expands that component
#|   of the list so that you can also see its children.
#| echo: false
#| out-width: NULL
knitr::include_graphics("screenshots/View-2.png", dpi = 220)
```

```{r}
#| label: fig-view-expand-2
#| fig.cap: >
#|   You can repeat this operation as many times as needed to get to the 
#|   data you're interested in. Note the bottom-right corner: if you click
#|   an element of the list, RStudio will give you the subsetting code
#|   needed to access it, in this case `x4[[2]][[2]][[2]]`.
#| echo: false
#| out-width: NULL
knitr::include_graphics("screenshots/View-3.png", dpi = 220)
```

### List-columns

Lists can also live inside a tibble, where we call them list-columns.
List-columns are useful because they allow you to shoehorn in objects that wouldn't wouldn't usually belong in a data frame.
List-columns are are used a lot in the tidymodels ecosystem, because it allows you to store things like models or resamples in a data frame.

Here's a simple example of a list-column:

```{r}
df <- tibble(
  x = 1:2, 
  y = c("a", "b"),
  z = list(list(1, 2), list(3, 4, 5))
)
df
```

There's nothing special about lists in a tibble; they behave like any other column:

```{r}
df |> 
  filter(x == 1)
```

Computing with them is harder, but that's because computing with lists is a harder; we'll come back to that in @sec-iteration.

The default print method just displays a rough summary of the contents.
The list column could be arbitrarily complex, so there's no good way to print it.
If you want to see it, you'll need to pull the list-column out and apply of the techniques that you learned above:

```{r}
df |> 
  filter(x == 1) |> 
  pull(z) |> 
  str()
```

Similarly, if you `View()` a data frame in RStudio, you'll get the standard tabular view, which doesn't allow you to selectively expand list columns.
To explore those fields you'll need to `pull()` and view, e.g.
`View(pull(df, z))`.

::: callout-note
## Base R

It's possible to put a list in a column of a `data.frame`, but it's a lot fiddlier.
However, base R doesn't make it easy to create list-columns because `data.frame()` treats a list as a list of columns:

```{r}
data.frame(x = list(1:3, 3:5))
```

You can prevent `data.frame()` from doing this with `I()`, but the result doesn't print particularly informatively:

```{r}
data.frame(
  x = I(list(1:3, 3:5)), 
  y = c("1, 2", "3, 4, 5")
)
```

Tibbles make it easier to work with list-columns because `tibble()` doesn't modify its inputs and the print method is designed with lists in mind.
:::

## Unnesting

Now that you've learned the basics of lists and list-columns, lets explore how you can turn them back into regular rows and columns.
We'll start with very simple sample data so you can get the basic idea, and then in the next section switch to more realistic examples.

List-columns tend to come in two basic forms: named and unnamed.
When the children are **named**, they tend to have the same names in every row.
When the children are **unnamed**, the number of elements tends to vary from row-to-row.
The following code creates an example of each.
In `df1`, every element of list-column `y` has two elements named `a` and `b`.
If `df2`, the elements of list-column `y` are unnamed and vary in length.

```{r}
df1 <- tribble(
  ~x, ~y,
  1, list(a = 11, b = 12),
  2, list(a = 21, b = 22),
  3, list(a = 31, b = 32),
)

df2 <- tribble(
  ~x, ~y,
  1, list(11, 12, 13),
  2, list(21),
  3, list(31, 32),
)
```

Named list-columns naturally unnest into columns: each named element becomes a new named column.
Unnamed list-columns naturally unnested in to rows: you'll get one row for each child.
tidyr provides two functions for these two case: `unnest_wider()` and `unnest_longer()`.
The following sections explain how they work.

### `unnest_wider()`

When each row has the same number of elements with the same names, like `df1`, it's natural to put each component into its own column with `unnest_wider()`:

```{r}
df1 |> 
  unnest_wider(y)
```

By default, the names of the new columns come exclusively from the names of the list, but you can use the `names_sep` argument to request that they combine the column name and the list names.
This is useful for disambiguating repeated names.

```{r}
df1 |> 
  unnest_wider(y, names_sep = "_")
```

We can also use `unnest_wider()` with unnamed list-columns, as in `df2`.
Since columns require names but the list lacks them, `unnest_wider()` will label them with consecutive integers:

```{r}
df2 |> 
  unnest_wider(y, names_sep = "_")
```

You'll notice that `unnested_wider()`, much like `pivot_wider()`, turns implicit missing values in to explicit missing values.

### `unnest_longer()`

When each row contains an unnamed list, it's most natural to put each element into its own row with `unnest_longer()`:

```{r}
df2 |> 
  unnest_longer(y)
```

Note how `x` is duplicated for each element inside of `y`: we get one row of output for each element inside the list-column.
But what happens if the list-column is empty, as in the following example?

```{r}
df6 <- tribble(
  ~x, ~y,
  "a", list(1, 2),
  "b", list(3),
  "c", list()
)
df6 |> unnest_longer(y)
```

We get zero rows in the output, so the row effectively disappears.
Once <https://github.com/tidyverse/tidyr/issues/1339> is fixed, you'll be able to keep this row, replacing `y` with `NA` by setting `keep_empty = TRUE`.

You can also unnest named list-columns, like `df1$y` into the rows.
Because the elements are named, and those names might be useful data, puts them in a new column with the suffix`_id`:

```{r}
df1 |> 
  unnest_longer(y)
```

If you don't want these `ids`, you can suppress this with `indices_include = FALSE`.
On the other hand, it's sometimes useful to retain the position of unnamed elements in unnamed list-columns.
You can do this with `indices_include = TRUE`:

```{r}
df2 |> 
  unnest_longer(y, indices_include = TRUE)
```

### Inconsistent types

What happens if you unnest a list-column contains different types of vector?
For example, take the following dataset where the list-column `y` contains two numbers, a factor, and a logical, which can't normally be mixed in a single column.

```{r}
df4 <- tribble(
  ~x, ~y,
  "a", list(1, "a"),
  "b", list(TRUE, factor("a"), 5)
)
```

`unnest_longer()` always keeps the set of columns change, while changing the number of rows.
So what happens?
How does `unnest_longer()` produce five rows while keeping everything in `y`?

```{r}
df4 |> 
  unnest_longer(y)
```

As you can see, the output contains a list-column, but every element of the list-column contains a single element.
Because `unnest_longer()` can't find a common type of vector, it keeps the original types in a list-column.
You might wonder if this breaks the commandment that every element of a column must be the same type --- not quite, because every element is a still a list, and each component of that list contains something different.

What happens if you find this problem in a dataset you're trying to rectangle?
I think there are two basic options.
You could use the `transform` argument to coerce all inputs to a common type.
It's not particularly useful here because there's only really one class that these five class can be converted to: character.

```{r}
df4 |> 
  unnest_longer(y, transform = as.character)
```

Another option would be to filter down to the rows that have values of a specific type:

```{r}
df4 |> 
  unnest_longer(y) |> 
  rowwise() |> 
  filter(is.numeric(y))
```

Then you can call `unnest_longer()` once more:

```{r}
df4 |> 
  unnest_longer(y) |> 
  rowwise() |> 
  filter(is.numeric(y)) |> 
  unnest_longer(y)
```

### Other functions

tidyr has a few other useful rectangling functions that we're not going to cover in this book:

-   `unnest_auto()` automatically picks between `unnest_longer()` and `unnest_wider()` based on the structure of the list-column. It's a great for rapid exploration, but I think it's ultimately a bad idea because it doesn't force you to understand how your data is structured, and makes your code harder to understand.
-   `unnest()` expands both rows and columns. It's useful when you have a list-column that contains a 2d structure like a data frame, which we don't see in this book.
-   `hoist()` allows you to reach into a deeply nested list and extract just the components that you need. It's mostly equivalent to repeated invocations of `unnest_wider()` + `select()` so you read up on it if you're trying to extract just a couple of important variables embedded in a bunch of data that you don't care about.

### Exercises

1.  From time-to-time you encounter data frames with multiple list-columns with aligned values.
    For example, in the following data frame, the values of `y` and `z` are aligned (i.e. `y` and `z` will always have the same length within a row, and the first value of `y` corresponds to the first value of `z`).
    What happens if you apply two `unnest_longer()` calls to this data frame?
    How can you preserve the relationship between `x` and `y`?
    (Hint: carefully read the docs).

    ```{r}
    df4 <- tribble(
      ~x, ~y, ~z,
      "a", list("y-a-1", "y-a-2"), list("z-a-1", "z-a-2"),
      "b", list("y-b-1", "y-b-2", "y-b-3"), list("z-b-1", "z-b-2", "z-b-3")
    )
    ```

## Case studies

So far you've learned about the simplest case of list-columns, where you need only a single call to `unnest_longer()` or `unnest_wider()`.
The main difference between real data and these simple examples, is with real data you'll see multiple levels of nesting.
For example, you might see named list nested inside an unnested list, or an unnamed list nested inside of another unnamed list nested inside a named list.
To handle these case you'll need to chain together multiple calls to `unnest_wider()` and/or `unnest_longer()`.

This section will work through some real rectangling challenges using datasets from the repurrrsive package that are inspired by datasets that we've encountered in the wild.
These challenges share the common feature that they're mostly just a sequence of multiple `unnest_wider()` and/or `unnest_longer()` calls, with a dash of dplyr where needed.

### Very wide data

We'll start by exploring `gh_repos` which contains data about some GitHub repositories retrived from the GitHub API. It's a very deeply nested list so it's to show the structure in this book; you might want to explore a little on your own with `View(gh_repos)` before we continue.

`gh_repos` is a list, but our tools work with list-columns, so we'll begin by putting it a tibble.
I call the column call `json` for reasons we'll get to later.

```{r}
repos <- tibble(json = gh_repos)
repos
```

This tibble contains 6 rows, one row for each child of `gh_repos`.
Each row contains a unnamed list with either 26 or 30 rows.
Since these are unnamed, we'll start with an `unnest_longer()` to put each child in its own row:

```{r}
repos |> 
  unnest_longer(json)
```

At first glance, it might seem like we haven't improved the situation: while we have more rows (176 instead of 6) each element of `json` is still a list.
However, there's an important difference: now each element is a **named** list so we can use `unnamed_wider()` to put each element into its own column:

```{r}
repos |> 
  unnest_longer(json) |> 
  unnest_wider(json) 
```

This has worked but the result is a little overwhelming: there are so many columns that tibble doesn't even print all of them!
We can see them all with `names()`:

```{r}
repos |> 
  unnest_longer(json) |> 
  unnest_wider(json) |> 
  names()
```

Let's select a few that look interesting:

```{r}
repos |> 
  unnest_longer(json) |> 
  unnest_wider(json) |> 
  select(id, full_name, owner, description)
```

You can use this to work back to understand `gh_repos`: each child was a GitHub user containing a list of up to 30 GitHub repositories that they created.

`owner` is another list-column, and since it a contains named list, we can use `unnest_wider()` to get at the values:

```{r}
#| error: true
repos |> 
  unnest_longer(json) |> 
  unnest_wider(json) |> 
  select(id, full_name, owner, description) |> 
  unnest_wider(owner)
```

Uh oh, this list column also contains an `id` column and we can't have two `id` columns in the same data frame.
Rather than following the advice to use `names_repair` (which would also work), I'll instead use `names_sep`:

```{r}
repos |> 
  unnest_longer(json) |> 
  unnest_wider(json) |> 
  select(id, full_name, owner, description) |> 
  unnest_wider(owner, names_sep = "_")
```

This gives another wide dataset, but you can see that `owner` appears to contain a lot of additional data about the person who "owns" the repository.

### Relational data

When you get nested data, it's not uncommon for it to contain data that we'd normally spread out into multiple data frames.
Take `got_chars`, for example.
Like `gh_repos` it's a list, so we start by turning it into a list-column of a tibble:

```{r}
chars <- tibble(json = got_chars)
chars
```

The `json` column contains named values, so we'll start by widening it:

```{r}
chars |> 
  unnest_wider(json)
```

And selecting a few columns just to make it easier to read:

```{r}
characters <- chars |> 
  unnest_wider(json) |> 
  select(id, name, gender, culture, born, died, alive)
characters
```

There are also many list-columns:

```{r}
chars |> 
  unnest_wider(json) |> 
  select(id, where(is.list))
```

Lets explore the `titles` column.
It's an unnamed list-column, so we'll unnest it into rows:

```{r}
chars |> 
  unnest_wider(json) |> 
  select(id, titles) |> 
  unnest_longer(titles)
```

You might expect to see this data in its own table because you could then join back to the characters data as needed.
To make this table I'll do a little cleaning; removing the rows contain empty strings and renaming `titles` to `title` since each row now only contains a single title.

```{r}
titles <- chars |> 
  unnest_wider(json) |> 
  select(id, titles) |> 
  unnest_longer(titles) |> 
  filter(titles != "") |> 
  rename(title = titles)
titles
```

Now, for example, we could use this table to all the characters that are captains and see all their titles:

```{r}
captains <- titles |> filter(str_detect(title, "Captain"))
captains

characters |> 
  semi_join(captains) |> 
  select(id, name) |> 
  left_join(titles)
```

You could imagine creating a table like this for each of the list-columns, then using joins to combine them with the character data as you need it.

### A dash of text analysis

What if we wanted to find the most common words in the title?
There are plenty of sophisticated ways to do this, but one simple way starts by using `str_split()` to break each element of `title` up into words by spitting on `" "`:

```{r}
titles |> 
  mutate(word = str_split(title, " "), .keep = "unused")
```

This creates a unnamed variable length list-column, so we can use `unnest_longer()`:

```{r}
titles |> 
  mutate(word = str_split(title, " "), .keep = "unused") |> 
  unnest_longer(word)
```

And then we can count that column to find the most common:

```{r}
titles |> 
  mutate(word = str_split(title, " "), .keep = "unused") |> 
  unnest_longer(word) |> 
  count(word, sort = TRUE)
```

Some of those words are not very interesting so we could create a list of common words to drop.
In text analysis this is commonly called stop words.

```{r}
stop_words <- tribble(
  ~ word,
  "of",
  "the"
)

titles |> 
  mutate(word = str_split(title, " "), .keep = "unused") |> 
  unnest_longer(word) |> 
  anti_join(stop_words) |> 
  count(word, sort = TRUE)
```

Breaking up text into individual fragments is a powerful idea that underlies much of text analysis.
If this sounds interesting, I'd recommend reading [Text Mining with R](https://www.tidytextmining.com) by Julia Silge and David Robinson.

### Deeply nested

We'll finish off this case studies with a list-column that's very deeply nested and requires repeated rounds of `unnest_wider()` and `unnest_longer()` to unravel: `gmaps_cities`.
This is a two column tibble containing five city names and the results of using Google's [geocoding API](https://developers.google.com/maps/documentation/geocoding) to determine their location:

```{r}
gmaps_cities
```

`json` is list-column with internal names, so we start with an `unnest_wider()`:

```{r}
gmaps_cities |> 
  unnest_wider(json)
```

This gives us the `status` and the `results`.
We'll drop the status column since they're all `OK`; in a real analysis, you'd also want capture all the rows where `status != "OK"` and figure out what went wrong.
`results` is an unnamed list, with either one or two elements (we'll see why shortly) so we'll unnest it into rows:

```{r}
gmaps_cities |> 
  unnest_wider(json) |> 
  select(-status) |> 
  unnest_longer(results)
```

Now `results` is a named list, so we'll use `unnest_wider()`:

```{r}
locations <- gmaps_cities |> 
  unnest_wider(json) |> 
  select(-status) |> 
  unnest_longer(results) |> 
  unnest_wider(results)
locations
```

Now we can see why two cities got two results: Washington matched both the Washington state and Washington, DC, and Arlington matched Arlington, Virginia and Arlington, Texas.

There are few different places we could go from here.
We might want to determine the exact location of the match, which is stored in the `geometry` list-column:

```{r}
locations |> 
  select(city, formatted_address, geometry) |> 
  unnest_wider(geometry)
```

That gives us new `bounds` (which gives a rectangular region) and the midpoint in `location`, which we can unnest to get latitude (`lat`) and longitude (`lng`):

```{r}
locations |> 
  select(city, formatted_address, geometry) |> 
  unnest_wider(geometry) |> 
  unnest_wider(location)
```

Extracting the bounds requires a few more steps

```{r}
locations |> 
  select(city, formatted_address, geometry) |> 
  unnest_wider(geometry) |> 
  # focus on the variables of interest
  select(!location:viewport) |>
  unnest_wider(bounds)
```

I then rename `southwest` and `northeast` (the corners of the rectangle) so I can use `names_sep` to create short but evocative names:

```{r}
locations |> 
  select(city, formatted_address, geometry) |> 
  unnest_wider(geometry) |> 
  select(!location:viewport) |>
  unnest_wider(bounds) |> 
  rename(ne = northeast, sw = southwest) |> 
  unnest_wider(c(ne, sw), names_sep = "_") 
```

Note that I unnest the two columns simultaneously by supplying a vector of variable names to `unnest_wider()`.

This one place where `hoist()`, mentioned briefly above, can be useful.
Once you've discovered the path to get to the components you're interested in, you can extract them directly using `hoist()`:

```{r}
locations |> 
  select(city, formatted_address, geometry) |> 
  hoist(
    geometry,
    ne_lat = c("bounds", "northeast", "lat"),
    sw_lat = c("bounds", "southwest", "lat"),
    ne_lng = c("bounds", "northeast", "lng"),
    sw_lng = c("bounds", "southwest", "lng"),
  )
```

If these case studies have whetted your appetite for more real-life rectangling, you can see a few more examples in `vignette("rectangling", package = "tidyr")`.

### Exercises

1.  Roughly estimate when `gh_repos` was created.
    Why can you only roughly estimate the date?

2.  The `owner` column of `gh_repo` contains a lot of duplicated information because each owner can have many repos.
    Can you construct a `owners` data frame that contains one row for each owner?
    (Hint: does `distinct()` work with `list-cols`?)

3.  Explain the following code line-by-line.
    Why is it interesting?
    Why does it work for this dataset but might not work in general?

    ```{r}
    tibble(json = got_chars) |> 
      unnest_wider(json) |> 
      select(id, where(is.list)) %>% 
      pivot_longer(
        where(is.list), 
        names_to = "name", 
        values_to = "value"
      ) %>% 
      unnest_longer(value)
    ```

4.  In `gmaps_cities`, what does `address_components` contain?
    Why does the length vary between rows?
    Unnest it appropriately to figure it out.
    (Hint: `types` always appears to contain two elements. Does `unnest_wider()` make it easier to work with than `unnest_longer()`?)
    .

## JSON

All of the case studies in the previous section came originally as JSON, one of the most common sources of hierarchical data.
In this section, you'll learn more about JSON and some common problems you might have.
JSON, short for javascript object notation, is a data format that grew out of the javascript programming language and has become an extremely common way of representing data.

``` json
{
  "name1": "value1",
  "name2": "value2"
}
```

Which in R you might represent as:

```{r}
list(
  name1 = "value1",
  name2 = "value2"
)
```

There are five types of things that JSON can represent

``` json
{
  "strings": "are surrounded by double doubles",
  "numbers": 123456,
  "boolean": [false, true],
  "arrays": [1, 2, 3, 4, 5],
  "objects": {
    "name1": "value1",
    "name2": "value2"
  },
  "null": null
}
```

You'll notice that these types don't embrace many of the types you've learned earlier in the book like factors, and date-times.
This is important: typically these data types will be encoded as string, and you'll need coerce to the correct data type.

Most of the time you won't deal with JSON directly, instead you'll use the jsonlite package, by Jeroen Oooms, to load it into R as a nested list.

### Data frames

JSON doesn't have any 2-dimension data structures, so how would you represent a data frame?

```{r}
df <- tribble(
  ~x, ~y,
  "a", 10,
  "x", 3
)
```

There are two ways: you can either make an struct of arrays, or an array of structs.

``` json
{
  "x": ["a", "x"],
  "y": [10, 3]
}
```

``` {.json .josn}
[
  {"x": "a", "y": 10},
  {"x": "x", "y": 3}
]
```

```{r}
df_col <- jsonlite::fromJSON('
  {
    "x": ["a", "x"],
    "y": [10, 3]
  }
')
tibble(json = list(df_col)) |> 
  unnest_wider(json) |> 
  unnest_longer(everything())
```

```{r}
df_row <- jsonlite::fromJSON(simplifyVector = FALSE, '
  [
    {"x": "a", "y": 10},
    {"x": "x", "y": 3}
  ]
')
tibble(json = list(df_row)) |> 
  unnest_longer(json) |> 
  unnest_wider(json)
```

Note that we have to wrap it in a `list()` because we have a single "thing" to unnest.