r4ds/import.Rmd

---
layout: default
title: Data import
---

# Data import

```{r, include = FALSE}
library(dplyr)
library(readr)
```

## Overview

You can't apply any of the tools you've applied so far to your own work, unless you can get your own data into R. In this chapter, you'll learn how to import:

* Flat files (like csv) with readr.
* Database queries with DBI.
* Data from web APIs with httr.
* Binary file formats (like excel or sas), with haven and readxl.

The common link between all these packages is they all aim to take your data and turn it into a data frame in R, so you can tidy it and then analyse it.

## Flat files

There are many ways to read flat files into R. If you've be using R for a while, you might be familiar with `read.csv()`, `read.fwf()` and friends. We're not going to use these base functions. Instead we're going to use `read_csv()`, `read_fwf()`, and friends from the readr package. Because:

* These functions are typically much faster (~10x) than the base equivalents.
  Long run running jobs also have a progress bar, so you can see what's
  happening. (If you're looking for raw speed, try `data.table::fread()`, 
  it's slightly less flexible than readr, but can be twice as fast.)
  
* They have more flexible parsers: they can read in dates, times, currencies,
  percentages, and more. 
  
* They fail to do some annoying things like converting character vectors to 
  factors, munging the column headers to make sure they're valid R 
  variable names, and using row names.

* They return objects with class `tbl_df`. As you saw in the dplyr chapter,
  this provides a nicer printing method, so it's easier to work with large
  datasets.

* They're designed to be as reproducible as possible - this means that you
  sometimes need to supply a few more arguments when using them the first
  time, but they'll definitely work on other peoples computers. The base R
  functions take a number of settings from system defaults, which means that
  code that works on your computer might not work on someone elses.

Make sure you have the readr package (`install.packages("readr")`).

Most of readr's functions are concerned with turning flat files into data frames:

* `read_csv()` read comma delimited files, `read_csv2()` reads semi-colon
  separated files (common in countries where `,` is used as the decimal place),
  `read_tsv()` reads tab delimited files, and `read_delim()` reads in files
  with a user supplied delimiter.

* `read_fwf()` reads fixed width files. You can specify fields either by their
  widths with `fwf_widths()` or theirs position with `fwf_positions()`. 
  `read_table()` reads a common variation of fixed width files where columns
  are separated by white space.

* `read_log()` reads Apache style logs. (But also check out
  [webreadr](https://github.com/Ironholds/webreadr) which is built on top 
  of `read_log()`, but provides many more helpful tools.)

readr also provides a number of functions for reading files off disk into simpler data structures:

* `read_file()` reads an entire file into a single string.

* `read_lines()` reads a file into a character vector with one element per line.

These might be useful for other programming tasks.

As well as reading data frame disk, readr also provides tools for working with data frames and character vectors in R:

* `type_convert()` applies the same parsing heuristics to the character columns
  in a data frame. You can override its choices using `col_types`.
  
For the rest of this chapter we'll focus on `read_csv()`. If you understand how to use this function, it will be straightforward to your knowledge to all the other functions in readr.

### Basics

The first two arguments of `read_csv()` are:

* `file`: path (or URL) to the file you want to load. Readr can automatically 
  decompress files ending in `.zip`, `.gz`, `.bz2`, and `.xz`. This can also
  be a literal csv file, which is useful for experimenting and creating
  reproducible examples.
  
* `col_names`: column names. There are three options:
  
    * `TRUE` (the default), which reads column names from the first row 
      of the file
      
    * `FALSE` number columns sequentially from `X1` to `Xn`.
    
    * A character vector, used as column names. If these don't match up
      with the columns in the data, you'll get a warning message.

EXAMPLE

### Column types

Readr uses a heuristic to figure out the types of your columns: it reads the first 1000 rows and uses some (moderately conservative) heuristics to figure out the type of each column. This is fast, and fairly robust. If readr detects the wrong type of data, you'll get warning messages. Readr prints out the first five, and you can access them all with `problems()`:

EXAMPLE

Typically, you'll see a lot of warnings if readr has guessed the column type incorrectly. This most often occurs when the first 1000 rows are different to the rest of the data. Perhaps there are a lot of missing data there, or maybe your data is mostly numeric but a few rows have characters. Fortunately, it's easy to fix these problems using the `col_type` argument.

(Note that if you have a very large file, you might want to set `n_max` to 10,000 or 100,000. That will speed up iteration while you're finding common problems)

Specifying the `col_type` looks like this:

```{r, eval = FALSE}
read_csv("mypath.csv", col_types = col(
  x = col_integer(),
  treatment = col_character()
))
```

You can use the following types of columns

* `col_integer()` (i) and `col_double()` (d) specify integer and doubles. 
  `col_logical()` (l) parses TRUE, T, FALSE and F into a logical vector.
  `col_character()` (c) leaves strings as is. 

* `col_number()` (n) is a more flexible parsed for numbers embedded in other 
  strings. It will look for the first number in a string, ignoring non-numeric 
  prefixes and suffixes. It will also ignoring the grouping mark specified by 
  the locale (see below for more details).
  
* `col_factor()` (f) allows you to load data directly into a factor if you know 
  what the levels are.
  
* `col_skip()` (_, -) completely ignores a column.

* `col_date()` (D), `col_datetime()` (T) and `col_time()` (t) parse into dates, 
  date times, and times as described below.

You might have noticed that each column parser has a one letter abbreviation, which you can instead of the full function call (assuming you're happy with the default arguments):

```{r, eval = FALSE}
read_csv("mypath.csv", col_types = cols(
  x = "i",
  treatment = "c"
))
```

(If you just have a few columns you supply a single string giving the type for each column: `i__dc`. See the documentation for more details. It's not as easy to understand as the `cols()` specification, so I'm not going to describe it further here.)

By default, any column not mentioned in `cols` will be guessed. If you'd rather those columns are simply not read in, use `cols_only()`. In that case, you can use `col_guess()` (?) if you want to guess the type of a column.

Each `col_XYZ()` function also has a corresponding `parse_XYZ()` that you can use on a character vector. This makes it easier to explore what each of the parsers does interactively.

```{r}
parse_integer(c("1", "2", "3"))
parse_logical(c("TRUE", "FALSE", "NA"))
parse_number(c("$1000", "20%", "3,000"))
parse_number(c("$1000", "20%", "3,000"))
```

Parsing occurs after leading and trailing whitespace has been removed (if not overridden with `trim_ws = FALSE`) and missing values listed in `na` have been removed:

```{r}
parse_logical(c("TRUE ", " ."), na = ".")
```

#### Datetimes

Readr provides three options depending on where you want a date (the number of days since 1970-01-01), a date time (the number of seconds since midnight 1970-01-01), or a time (i.e. the number of seconds since midnight). The defaults read:

* Date times: an [ISO8601](https://en.wikipedia.org/wiki/ISO_8601) date time.
* Date: a year, optional separator, month, optional separator, day.
* Time: an hour, optional colon, hour, optional colon, minute, optional colon,
  optional seconds, optional am/pm.

```{r}
parse_datetime("2010-10-01T2010")
parse_date("2010-10-01")
parse_time("20:10:01")
```

If these defaults don't work for your data you can supply your own date time formats, built up of the following pieces:

* Year: `%Y` (4 digits). `%y` (2 digits); 00-69 -> 2000-2069, 70-99 -> 1970-1999.

* Month: `%m` (2 digits), `%b` (abbreviated name), `%B` (full name).

* Day: `%d` (2 digits), `%e` (optional leading space).

* Hour: `%H`.

* Minutes: `%M`.

* Seconds: `%S` (integer seconds), `%OS` (partial seconds).

* Time zone: `%Z` (as name, e.g. `America/Chicago`), `%z` (as offset from UTC, 
  e.g. `+0800`). If you're American, note that "EST" is a Canadian time zone 
  that does not have daylight savings time. It is \emph{not} Eastern Standard 
  Time!

* AM/PM indicator: `%p`.

* Non-digits: `%.` skips one non-digit charcter, `%*` skips any number of 
  non-digits.

The best way to figure out the correct string is to create a few examples in a character vector, and test with one of the parsing functions. For example:

```{r}
parse_date("01/02/15", "%m/%d/%y")
parse_date("01/02/15", "%d/%m/%y")
parse_date("01/02/15", "%y/%m/%d")
```

Then when you read in the data with `read_csv()` you can easily translate to the `col_date()` format.

### International data

The goal of readr's locales is to encapsulate the common options that vary between languages and different regions of the world. This includes:

* Names of months and days, used when parsing dates.
* The default time zones, used when parsing date times.
* The character encoding, used when reading non-ASCII strings.
* Default date and time formats, used when guessing column types.
* The decimal and grouping marks, used when reading numbers.

Readr is designed to be independent of your current locale settings. This makes a bit more hassle in the short term, but makes it much much easier to share your code with others: if your readr code works locally, it will also work for everyone else in the world. The same is not true for base R code, since it often inherits defaults from your system settings. Just because data ingest code works for you doesn't mean that it will work for someone else in another country.

The settings you are most like to need to change are:

*   The names of days and months:

    ```{r}
    locale("fr")
    locale("fr", asciify = TRUE)
    ```
    
*   The character encoding used in the file.  If you don't know the encoding
    you can use `guess_encoding()`. It's not perfect, but if you have a decent
    sample of text, it's likely to be able to figure it out.
    
    Readr converts all strings into UTF-8 as this is safest to work with across
    platforms. (It's also what every stringr operation does.)

### Exercises

* Parse these dates (incl. non-English examples).
* Parse these example files.
* Parse this fixed width file.

## Databases

## Web APIs

## Binary files

Needs to discuss how data types in different languages are converted to R. Similarly for missing values.


## Tibble diffs

`data_frame()` is a nice way to create data frames. It encapsulates best practices for data frames:

  * It never changes an input's type (i.e., no more `stringsAsFactors = FALSE`!).
    
    ```{r}
    data.frame(x = letters) %>% sapply(class)
    data_frame(x = letters) %>% sapply(class)
    ```
    
    This makes it easier to use with list-columns:
    
    ```{r}
    data_frame(x = 1:3, y = list(1:5, 1:10, 1:20))
    ```
    
    List-columns are most commonly created by `do()`, but they can be useful to
    create by hand.
      
  * It never adjusts the names of variables:
  
    ```{r}
    data.frame(`crazy name` = 1) %>% names()
    data_frame(`crazy name` = 1) %>% names()
    ```

  * It evaluates its arguments lazily and sequentially:
  
    ```{r}
    data_frame(x = 1:5, y = x ^ 2)
    ```

  * It adds the `tbl_df()` class to the output so that if you accidentally print a large 
    data frame you only get the first few rows.
    
    ```{r}
    data_frame(x = 1:5) %>% class()
    ```

  * It changes the behaviour of `[` to always return the same type of object:
    subsetting using `[` always returns a `tbl_df()` object; subsetting using 
    `[[` always returns a column.
    
    You should be aware of one case where subsetting a `tbl_df()` object  
    will produce a different result than a `data.frame()` object:
  
    ```{r}
    df <- data.frame(a = 1:2, b = 1:2)
    str(df[, "a"])
    
    tbldf <- tbl_df(df)
    str(tbldf[, "a"])
    ```
    
  * It never uses `row.names()`. The whole point of tidy data is to 
    store variables in a consistent way. So it never stores a variable as 
    special attribute.
  
  * It only recycles vectors of length 1. This is because recycling vectors of greater lengths 
    is a frequent source of bugs.

### Coercion

To complement `data_frame()`, dplyr provides `as_data_frame()` to coerce lists into data frames. It does two things:

* It checks that the input list is valid for a data frame, i.e. that each element
  is named, is a 1d atomic vector or list, and all elements have the same 
  length.
  
* It sets the class and attributes of the list to make it behave like a data frame.
  This modification does not require a deep copy of the input list, so it's
  very fast.
  
This is much simpler than `as.data.frame()`. It's hard to explain precisely what `as.data.frame()` does, but it's similar to `do.call(cbind, lapply(x, data.frame))` - i.e. it coerces each component to a data frame and then `cbinds()` them all together. Consequently `as_data_frame()` is much faster than `as.data.frame()`:

```{r}
l2 <- replicate(26, sample(100), simplify = FALSE)
names(l2) <- letters
microbenchmark::microbenchmark(
  as_data_frame(l2),
  as.data.frame(l2)
)
```

The speed of `as.data.frame()` is not usually a bottleneck when used interactively, but can be a problem when combining thousands of messy inputs into one tidy data frame.

### tbl_dfs vs data.frames

There are three key differences between tbl_dfs and data.frames:

*   When you print a tbl_df, it only shows the first ten rows and all the
    columns that fit on one screen. It also prints an abbreviated description
    of the column type:
    
    ```{r}
    data_frame(x = 1:1000)
    ```
    
    You can control the default appearance with options:
    
    * `options(dplyr.print_max = n, dplyr.print_min = m)`: if more than `n`
      rows print `m` rows. Use `options(dplyr.print_max = Inf)` to always
      show all rows.
    
    * `options(dply.width = Inf)` will always print all columns, regardless
       of the width of the screen.

    
*   When you subset a tbl\_df with `[`, it always returns another tbl\_df. 
    Contrast this with a data frame: sometimes `[` returns a data frame and
    sometimes it just returns a single column:
    
    ```{r}
    df1 <- data.frame(x = 1:3, y = 3:1)
    class(df1[, 1:2])
    class(df1[, 1])
    
    df2 <- data_frame(x = 1:3, y = 3:1)
    class(df2[, 1:2])
    class(df2[, 1])
    ```
    
    To extract a single column it's use `[[` or `$`:
    
    ```{r}
    class(df2[[1]])
    class(df2$x)
    ```

*   When you extract a variable with `$`, tbl\_dfs never do partial 
    matching. They'll throw an error if the column doesn't exist:
    
    ```{r, error = TRUE}
    df <- data.frame(abc = 1)
    df$a
    
    df2 <- data_frame(abc = 1)
    df2$a
    ```