Functional programming and iteration with purrr

# Functional programming and iteration with purrr
### 2020-08-22

---

background-image: url(http://hexb.in/hexagons/purrr.png)
background-position: 90% 26%

# purrr: A functional programming toolkit for R

## Complete and consistent set of tools for working with functions and vectors

---
class: inverse, middle

# Problems we want to solve:
1. Making code clear
2. Making code safe
3. Working with lists and data frames

---

# Lists, vectors, and data.frames (or tibbles)

```r
c(char = "hello", num = 1)
```

```
##    char     num 
## "hello"     "1"
```

---

# lists can contain any object

```r
list(char = "hello", num = 1, fun = mean)
```

```
## $char
## [1] "hello"
## 
## $num
## [1] 1
## 
## $fun
## function (x, ...) 
## UseMethod("mean")
## <bytecode: 0x7fb922834d08>
## <environment: namespace:base>
```

---

## Your Turn 1

```r
measurements <- list(
  blood_glucose = rnorm(10, mean = 140, sd = 10), 
  age = rnorm(5, mean = 40, sd = 5), 
  heartrate = rnorm(20, mean = 80, sd = 15)
)
```

### There are two ways to subset lists: dollar signs and brackets. Try to subset `blood_glucose` from `measurements` using these approaches. Are they different? What about `measurements[["blood_glucose"]]`?

---

## Your Turn 1

```r
measurements["blood_glucose"]
```

```
## $blood_glucose
##  [1] 127.9293 142.7743 150.8444 116.5430 144.2912 145.0606 134.2526 134.5337 134.3555 131.0996
```

```r
measurements$blood_glucose
```

```
##  [1] 127.9293 142.7743 150.8444 116.5430 144.2912 145.0606 134.2526 134.5337 134.3555 131.0996
```

```r
measurements[["blood_glucose"]]
```

```
##  [1] 127.9293 142.7743 150.8444 116.5430 144.2912 145.0606 134.2526 134.5337 134.3555 131.0996
```
]
---

#  data frames are lists

```r
x <- list(char = "hello", num = 1)
as.data.frame(x)
```

```
##    char num
## 1 hello   1
```

---

#  data frames are lists

```r
library(gapminder)
head(gapminder$pop)
```

```
## [1]  8425333  9240934 10267083 11537966 13079460 14880372
```

---

#  data frames are lists

```r
gapminder[1:6, "pop"]
```

---

#  data frames are lists

```r
gapminder[1:6, "pop"]
```

```
## # A tibble: 6 x 1
##        pop
##      <int>
## 1  8425333
## 2  9240934
## 3 10267083
## 4 11537966
## 5 13079460
## 6 14880372
```

---

#  data frames are lists

```r
head(gapminder[["pop"]])
```

```
## [1]  8425333  9240934 10267083 11537966 13079460 14880372
```

---

#  vectorized functions don't work on lists

```r
sum(rnorm(10))
```

---

#  vectorized functions don't work on lists

```r
sum(rnorm(10))
```

```
## [1] -3.831574
```

---

#  vectorized functions don't work on lists

```r
sum(rnorm(10))
```

```
## [1] -3.831574
```

```r
sum(list(x = rnorm(10), y = rnorm(10), z = rnorm(10)))
```

---

#  vectorized functions don't work on lists

```r
sum(rnorm(10))
```

```
## [1] -3.831574
```

```r
sum(list(x = rnorm(10), y = rnorm(10), z = rnorm(10)))
```

```
## Error in sum(list(x = rnorm(10), y = rnorm(10), z = rnorm(10))): invalid 'type' (list) of argument
```

---

background-image: url(http://hexb.in/hexagons/purrr.png)
background-position: 95% 2%

# map(.x, .f)

## .x: a vector, list, or data frame

## .f: a function

## Returns a list

---

# Using map()

```r
library(purrr)
x_list <- list(x = rnorm(10), y = rnorm(10), z = rnorm(10))

map(x_list, mean)
```

---

# Using map()

```r
library(purrr)
*x_list <- list(x = rnorm(10), y = rnorm(10), z = rnorm(10))

map(x_list, mean)
```

---

# Using map()

```r
library(purrr)
x_list <- list(x = rnorm(10), y = rnorm(10), z = rnorm(10))

*map(x_list, mean)
```

---

# Using map()

```r
library(purrr)
x_list <- list(x = rnorm(10), y = rnorm(10), z = rnorm(10))

map(x_list, mean)
```

```
## $x
## [1] -0.6097971
## 
## $y
## [1] -0.2788647
## 
## $z
## [1] 0.6165922
```

---

---

---

---

## Your Turn 2

### Read the code in the first chunk and predict what will happen

### Run the code in the first chunk. What does it return?

```r
list(
  sum_blood_glucose = sum(measurements$blood_glucose),
  sum_age = sum(measurements$age),
  sum_heartrate = sum(measurements$heartrate)
)
```

### Now, use `map()` to create the same output.

---

## Your Turn 2

```r
map(measurements, sum)
```

```
## $blood_glucose
## [1] 1361.684
## 
## $age
## [1] 193.8606
## 
## $heartrate
## [1] 1509.304
```

---

## using `map()` with data frames

```r
library(dplyr)
gapminder %>% 
  select(where(is.numeric)) %>%
  map(sd)
```

---

## using `map()` with data frames

```r
library(dplyr)
*gapminder %>%
* select(where(is.numeric)) %>%
  map(sd)
```

---

## using `map()` with data frames

```r
library(dplyr)
gapminder %>% 
  select(where(is.numeric)) %>%
* map(sd)
```

---

## using `map()` with data frames

```r
library(dplyr)
gapminder %>% 
  select(where(is.numeric)) %>%
  map(sd)
```

```
## $year
## [1] 17.26533
## 
## $lifeExp
## [1] 12.91711
## 
## $pop
## [1] 106157897
## 
## $gdpPercap
## [1] 9857.455
```

---

## Your Turn 3

### Pass diabetes to `map()` and map using `class()`. What are these results telling you?

---

## Your Turn 3

```r
head(
* map(diabetes, class),
  3
)
```

```
## $id
## [1] "numeric"
## 
## $chol
## [1] "numeric"
## 
## $stab.glu
## [1] "numeric"
```

---

#  Review: writing functions

```r
x <- x^2
x <- scale(x)
x <- max(x)
```

---

#  Review: writing functions

```r
x <- x^2
x <- scale(x)
x <- max(x)

y <- x^2
y <- scale(y)
y <- max(y)

z <- z^2
z <- scale(x)
z <- max(z)
```

---

#  Review: writing functions

```r
x <- x^2
x <- scale(x)
x <- max(x)

*y <- x^2
y <- scale(y)
y <- max(y)

z <- z^2
*z <- scale(x)
z <- max(z)
```

---

#  Review: writing functions

```r
*x <- x^3
x <- scale(x)
x <- max(x)

*y <- x^2
y <- scale(y)
y <- max(y)

*z <- z^2
z <- scale(x)
z <- max(z)
```

---

#  Review: writing functions

```r
.f <- function(x) {
  x <- x^3
  x <- scale(x)
  
  max(x)
}

.f(x)
.f(y)
.f(z)
```

---

# **If you copy and paste your code three times, it's time to write a function**

---

## Your Turn 4

### Write a function that returns the mean and standard deviation of a numeric vector.
### Give the function a name
### Find the mean and SD of `x`
### Map your function to `measurements`

---

## Your Turn 4

```r
mean_sd <- function(x) {
  x_mean <- mean(x)
  x_sd <- sd(x)
  tibble(mean = x_mean, sd = x_sd)
}
  
map(measurements, mean_sd)
```
---

## Your Turn 4

```
## $blood_glucose
## # A tibble: 1 x 2
##    mean    sd
##   <dbl> <dbl>
## 1  136.  9.96
## 
## $age
## # A tibble: 1 x 2
##    mean    sd
##   <dbl> <dbl>
## 1  38.8  3.91
## 
## $heartrate
## # A tibble: 1 x 2
##    mean    sd
##   <dbl> <dbl>
## 1  75.5  13.8
```

---

class: inverse
# Three ways to pass functions to `map()`
1. pass directly to `map()`
2. use an anonymous function
3. use ~

---

---

---

---

```r
map(gapminder, ~length(unique(.x)))
```

---

```r
map(gapminder, ~length(unique(.x)))
```

```
## $country
## [1] 142
## 
## $continent
## [1] 5
## 
## $year
## [1] 12
## 
## $lifeExp
## [1] 1626
## 
## $pop
## [1] 1704
## 
## $gdpPercap
## [1] 1704
```

---

# Returning types

---

# Returning types

```r
map_int(gapminder, ~length(unique(.x)))
```

---

# Returning types

```r
map_int(gapminder, ~length(unique(.x)))
```

```
##   country continent      year   lifeExp       pop gdpPercap 
##       142         5        12      1626      1704      1704
```

---

## Your Turn 5

### Do the same as #3 above but return a vector instead of a list.

---

## Your Turn 5
.small[

```r
map_chr(diabetes, class)
```

```
##          id        chol    stab.glu         hdl       ratio       glyhb    location         age      gender      height      weight       frame       bp.1s       bp.1d       bp.2s 
##   "numeric"   "numeric"   "numeric"   "numeric"   "numeric"   "numeric" "character"   "numeric" "character"   "numeric"   "numeric" "character"   "numeric"   "numeric"   "numeric" 
##       bp.2d       waist         hip    time.ppn 
##   "numeric"   "numeric"   "numeric"   "numeric"
```
]
---

## Your Turn 6

### Check `diabetes` for any missing data.

### Using the ~.f(.x) shorthand, check each column for any missing values using `is.na()` and `any()`
### Return a logical vector. Are any columns missing data? What happens if you don't include `any()`? Why?
### Try counting the number of missing, returning an integer vector

---

## Your Turn 6

```r
map_lgl(diabetes, ~any(is.na(.x)))
```

```
##       id     chol stab.glu      hdl    ratio    glyhb location      age   gender   height   weight    frame    bp.1s    bp.1d    bp.2s    bp.2d    waist      hip time.ppn 
##    FALSE     TRUE    FALSE     TRUE     TRUE     TRUE    FALSE    FALSE    FALSE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE
```
]
---

## Your Turn 6

```r
map_int(diabetes, ~sum(is.na(.x)))
```

```
##       id     chol stab.glu      hdl    ratio    glyhb location      age   gender   height   weight    frame    bp.1s    bp.1d    bp.2s    bp.2d    waist      hip time.ppn 
##        0        1        0        1        1       13        0        0        0        5        1       12        5        5      262      262        2        2        3
```
]
---

## Your Turn 7

### Turn `diabetes` into a list split by `location` using the `split()` function. Check its length.
### Fill in the `model_lm` function to model `chol` (the outcome) with `ratio` and pass the `.data` argument to `lm()`
### map `model_lm` to `diabetes_list` so that it returns a data frame (by row).

---

## Your Turn 7

```r
*diabetes_list <- split(diabetes, diabetes$location)
length(diabetes_list)
model_lm <- function(.data) {
* mdl <- lm(chol ~ ratio, data = .data)
  # get model statistics
  broom::glance(mdl)
}

*map(diabetes_list, model_lm)
```

---

## Your Turn 7
.small[

```
## [1] 2
```

```
## $Buckingham
## # A tibble: 1 x 12
##   r.squared adj.r.squared sigma statistic  p.value    df
##       <dbl>         <dbl> <dbl>     <dbl>    <dbl> <dbl>
## 1     0.252         0.248  38.8      66.4 4.11e-14     1
## # … with 6 more variables: logLik <dbl>, AIC <dbl>,
## #   BIC <dbl>, deviance <dbl>, df.residual <int>,
## #   nobs <int>
## 
## $Louisa
## # A tibble: 1 x 12
##   r.squared adj.r.squared sigma statistic  p.value    df
##       <dbl>         <dbl> <dbl>     <dbl>    <dbl> <dbl>
## 1     0.204         0.201  39.4      51.7 1.26e-11     1
## # … with 6 more variables: logLik <dbl>, AIC <dbl>,
## #   BIC <dbl>, deviance <dbl>, df.residual <int>,
## #   nobs <int>
```
]

---

# map2(.x, .y, .f)

## .x, .y: a vector, list, or data frame

## .f: a function

## Returns a list

---

---

---

---

# map2()

```r
means <- c(-3, 4, 2, 2.3)
sds <- c(.3, 4, 2, 1)
  
map2_dbl(means, sds, rnorm, n = 1)
```

---

# map2()

```r
means <- c(-3, 4, 2, 2.3)
sds <- c(.3, 4, 2, 1)
  
*map2_dbl(means, sds, rnorm, n = 1)
```

---

# map2()

```r
means <- c(-3, 4, 2, 2.3)
sds <- c(.3, 4, 2, 1)
  
map2_dbl(means, sds, rnorm, n = 1)
```

```
## [1] -2.997932  2.178125  1.266952  2.948287
```

---

## Your Turn 8

### Split the gapminder dataset into a list by country
### Create a list of models using `map()`. For the first argument, pass `gapminder_countries`. For the second, use the `~.f()` notation to write a model with `lm()`. Use `lifeExp` on the left hand side of the formula and `year` on the second. Pass `.x` to the `data` argument.
### Use `map2()` to take the models list and the data set list and map them to `predict()`. Since we're not adding new arguments, you don't need to use `~.f()`.

---

## Your Turn 8

```r
*gapminder_countries <- split(gapminder, gapminder$country)
models <- map(gapminder_countries, ~ lm(lifeExp ~ year, data = .x))
preds <- map2(models, gapminder_countries, predict)
head(preds, 3)
```

---

## Your Turn 8

```r
gapminder_countries <- split(gapminder, gapminder$country)
*models <- map(gapminder_countries, ~ lm(lifeExp ~ year, data = .x))
preds <- map2(models, gapminder_countries, predict)
head(preds, 3)
```

---

## Your Turn 8

```r
gapminder_countries <- split(gapminder, gapminder$country)
models <- map(gapminder_countries, ~ lm(lifeExp ~ year, data = .x))
*preds <- map2(models, gapminder_countries, predict)
head(preds, 3)
```

---

## Your Turn 8

```
## $Afghanistan
##        1        2        3        4        5        6 
## 29.90729 31.28394 32.66058 34.03722 35.41387 36.79051 
## 
## $Albania
##        1        2        3        4        5        6 
## 59.22913 60.90254 62.57596 64.24938 65.92279 67.59621 
## 
## $Algeria
##        1        2        3        4        5        6 
## 43.37497 46.22137 49.06777 51.91417 54.76057 57.60697
```

---

---

# Other mapping functions

## **pmap()** and friends: take n lists or data frame with argument names
---

# Other mapping functions

## ~~pmap() and friends: take n lists or data frame with argument names~~
## **walk()** and friends: for side effects like plotting; returns input invisibly
---

# Other mapping functions

## ~~pmap() and friends: take n lists or data frame with argument names~~
## ~~walk() and friends: for side effects like plotting; returns input invisibly~~
## **imap()** and friends: includes counter `i`
---

# Other mapping functions

## ~~pmap() and friends: take n lists or data frame with argument names~~
## ~~walk() and friends: for side effects like plotting; returns input invisibly~~
## ~~imap() and friends: includes counter `i`~~
## **map_if()**, **map_at()**: Apply only to certain elements

---

---

## Your turn 9

#### Create a new directory using the fs package. Call it "figures".

#### Write a function to plot a line plot of a given variable in gapminder over time, faceted by continent. Then, save the plot (how do you save a ggplot?). For the file name, paste together the folder, name of the variable, and extension so it follows the pattern `"folder/variable_name.png"`
#### Create a character vector that has the three variables we'll plot: "lifeExp", "pop", and "gdpPercap".
#### Use `walk()` to save a plot for each of the variables

---

## Your turn 9

```r
fs::dir_create("figures")

ggsave_gapminder <- function(variable) {
  #  we're using `aes_string()` so we don't need the curly-curly syntax
  p <- ggplot(
    gapminder, 
    aes_string(x = "year", y = variable, color = "country")
  ) + 
    geom_line() + 
    scale_color_manual(values = country_colors) + 
    facet_wrap(vars(continent.)) + 
    theme(legend.position = "none")
    
  ggsave(
    filename = paste0("figures/", variable, ".png"), 
    plot = p, 
    dpi = 320
  )
}
```

---

## Your turn 9

```r
vars <- c("lifeExp", "pop", "gdpPercap")
walk(vars, ggsave_gapminder)
```

---

# Base R

---

# Benefits of purrr
1. Consistent 
2. Type-safe
3. ~f(.x)

---

## Loops vs functional programming

```r
x <- rnorm(10)
y <- map(x, mean)
```

```r
x <- rnorm(10)
y <- vector("list", length(x))
for (i in seq_along(x)) {
  y[[i]] <- mean(x[[i]])
}
```

---

## Loops vs functional programming

```r
x <- rnorm(10)
y <- map(x, mean)
```

```r
x <- rnorm(10) 
*y <- vector("list", length(x))
for (i in seq_along(x)) {
  y[[i]] <- mean(x[[i]]) 
}
```

---

## Loops vs functional programming

```r
x <- rnorm(10)
y <- map(x, mean)
```

```r
x <- rnorm(10)
y <- vector("list", length(x)) 
*for (i in seq_along(x)) {
  y[[i]] <- mean(x[[i]]) 
}
```

---

## Loops vs functional programming

```r
x <- rnorm(10)
y <- map(x, mean)
```

```r
x <- rnorm(10)
y <- vector("list", length(x)) 
for (i in seq_along(x)) { 
* y[[i]] <- mean(x[[i]])
}
```

---

# **Of course someone has to write loops. It doesn’t have to be you.**
# **—Jenny Bryan**

---

# Working with lists and nested data

---

# Working with lists and nested data

---

# Adverbs: Modify function behavior

---

# Learn more!
## [Jenny Bryan's purrr tutorial](https://jennybc.github.io/purrr-tutorial/): A detailed introduction to purrr. Free online.
## [R for Data Science](http://r4ds.had.co.nz/): A comprehensive but friendly introduction to the tidyverse. Free online.
## [RStudio Primers](https://rstudio.cloud/learn/primers): Free interactive courses in the Tidyverse