Functional programming in R

class: center, middle

![:scale 30%](/assets/images/coding_club_logo_1.png)

<!-- Adjust the presentation to the session. Focus on the challenges,
 this is not a coding tutorial.

Note, to include figures, store the image in the `/docs/assets/images/yyyymmdd/`
    folder and use the jekyll base.url reference as done in this template
    or see https://jekyllrb.com/docs/liquid/tags/#links.
    using the scale attribute ![:scale 30%](...), you can adjust the image size.
-->

# 30 JUNE 2020

## INBO coding club

Everywhere you wanna be...

---
class: center, middle

![:scale 90%](/assets/images/20200630/20200630_functional_programming.png)

---
class: left, middle

- Functional programming: _"wrap up for loops in a function, and call that function instead of using the for loop directly"_, [R for Data Science](https://r4ds.had.co.nz/iteration.html#for-loops-vs.functionals)

- functional programming in basic R: `apply()`, `lapply()`, `tapply()`

- The `purrr` package provides a complete and consistent set of tools

<center>![:scale 30%](/assets/images/20200630/20200630_purrr_logo.png) </center>

Are purrr or apply() functions faster than loops! NO. It's just a question of clarity: **code is easier to write and to read**. But first you need a little of practice, and here we are: doing some purrr "fitness" exercises at the coding _gym_ club.

---
class: center, middle

![:scale 90%](/assets/images/20200630/20200630_cheatsheet_purrr.png)
[Download cheatsheet here](https://github.com/inbo/coding-club/blob/master/cheat_sheets/20181023_cheat_sheet_purrr.pdf)

---
class: center, middle

### How to get started?

Check the [Each session setup](https://inbo.github.io/coding-club/gettingstarted.html#each-session-setup) to get started.

### First time coding club?

Check the [First time setup](https://inbo.github.io/coding-club/gettingstarted.html#first-time-setup) section to setup.

---
class: left, middle

![:scale 100%](/assets/images/coding_club_sticky_concept.png)
 No yellow sticky notes online :-( We use hackmd (see next slide) but basic principle doesn't change.

---
class: center, middle

### Share your code during the coding session!

Go to https://hackmd.io/ZDc7gTTHRm-ZcjYq_2f2Fg?both

---
class: left, middle

# Install and load the required packages:

Copy-paste and run the following R script:

```r
# CRAN packages
installed <- rownames(installed.packages())
required <- c(
 "remotes",
 "tidyverse",
 "lubridate",
 "rprojroot"
 )
if (!all(required %in% installed)) {
 install.packages(required[!required %in% installed])
}

# non CRAN packages
if (!"palmerpenguins" %in% installed) {
  remotes::install_github("allisonhorst/palmerpenguins")
}

# load packages
library(tidyverse)
library(lubridate)
library(rprojroot)
library(palmerpenguins)
```

---
class: left, middle

# Download data and code

Today we will use an R script and some datasets. Download* the R script in your folder `src ` and the dataset in your folder `data`.

- R script: [20200630/20200630_challenges.R](https://github.com/inbo/coding-club/blob/master/src/20200630/20200630_challenges.R)
- Datasets with sensor data: [20200630/20200630_sensor_A1.txt](https://github.com/inbo/coding-club/blob/master/data/20200630/20200630_sensor_A1.txt), [20200630/20200630_sensor_G7.txt](https://github.com/inbo/coding-club/blob/master/data/20200630/20200630_sensor_G7.txt), and [20200630/20200630_sensor_H4.txt](https://github.com/inbo/coding-club/blob/master/data/20200630/20200630_sensor_H4.txt)

* __Note__: check the getting started instructions on [how to download a single file](https://inbo.github.io/coding-club/gettingstarted.html#each-session-setup)

---
class: left, middle

# Last month...

## Why loops?

[DRY principle](https://en.wikipedia.org/wiki/Don%27t_repeat_yourself) ("Don't Repeat Yourself")
![:scale 90%](/assets/images/20200528/20200528_loops_why.png)

DRY is the enemy of WET ("write every time", "write everything twice", "we enjoy typing" or "waste everyone's time")

---
class: left, middle

# Basics: `map()`

The _mantra_ of this 2 hours coding session is: **FOR EACH __ DO __**
![:scale 85%](/assets/images/20200630/20200630_map_basics.png)

---
class: left, middle

<center>![:scale 50%](/assets/images/20200630/20200630_map_basics.png)</center>

```r
# height of 5 giraffes
height_giraffe <- runif(n = 5, min = 4.3, max = 5.7)
```
- FOR EACH element of `height_giraffe` DO floor()
- FOR EACH element of `height_giraffe` DO as.character()

---
class: left, middle

<center>![:scale 50%](/assets/images/20200630/20200630_map_basics.png)</center>

```r
# weight of 5 giraffes
weight_giraffe <- rnorm(n = 5, mean = 1192, sd = 300)
# data.frame with height and weight
df_giraffe <- tibble(height = height_giraffe,
 weight = weight_giraffe)
```

- FOR EACH column of `df_giraffe` DO calculate mean
- FOR EACH column of `df_giraffe` DO calculate median

---
class: left, middle

<center>![:scale 50%](/assets/images/20200630/20200630_map_basics.png)</center>

```r
# nicknames of 5 giraffes
names_giraffe <- c("Damiano", "Dirk", "Emma", "Hans", "Joost")
# List with height and weight of 5 giraffes
info_giraffe <- transpose(df_giraffe) %>% set_names(names_giraffe)
# Add specific fields to some giraffes
info_giraffe$Dirk$subspecies <- "Giraffa camelopardalis subsp. antiquorum"
info_giraffe$Emma$GPS_trackerID <- "AFG4953"
```

- FOR EACH giraffe of `info_giraffe` DO get class
- FOR EACH giraffe of `info_giraffe` DO get length

---
background-image: url(/assets/images/background_challenge_1.png)
class: left, middle

# Challenge 1

Write `map()` statements to:

1. Compute the mean of every column in [`swiss`](https://stat.ethz.ch/R-manual/R-devel/library/datasets/html/swiss.html)
2. Determine the type of each column in [`iris`](https://stat.ethz.ch/R-manual/R-devel/library/datasets/html/iris.html)
3. Set swiss and iris as a named list of two data.frames
```r
dfs <- list(swiss = swiss, iris = iris)
```
and get the number of rows of each data.frame
4. `map` returns ALWAYS a list, **ALWAYS**. Apply the right `map_*()` variants to the previous exercises to return a numeric vector (1), a character vector (2) or a data.frame.

Hint: [tutorial at 21.5](https://r4ds.had.co.nz/iteration.html#the-map-functions) or cheat sheet

---
class: left, middle

# The `map` strategy

So far it is so plain. Data science is rarely so easy.
Example: <bold>FOR EACH</bold> giraffe of info_giraffe <bold>DO</bold> calculate the body mass index (weight / height^2)

```r
bmi <- function(weight, height) weight / ((height)^2)
```

But:

```r
map(info_giraffe, bmi)
Error in .f(.x[[i]], ...) : argument "height" is missing, with no default
```

---
class: left, middle

# The `map` strategy

1. Do it for one element
2. Turn it into a <bold>recipe</bold>
3. Use `map()` to do it for all elements

---
class: left, middle

# The `map` strategy. Step 1

## 1. Do it for one element

```r
Joost <- info_giraffe$Joost
bmi(Joost$weight, Joost$height)
```

---
class: left, middle

# The `map` strategy. Step 2

2.Turn it into a <bold>recipe</bold>

Make it a **formula** (`~`) and use `.x` as pronoun

```r
~ bmi(.x$weight, .x$height)
```

---
class: left, middle

# The `map` strategy. Step 3

3.Use `map()` to do it for all elements

```r
map(info_giraffe, ~ bmi(.x$weight, .x$height))
```

or if you don't feel confident with "~" and ".x":

```r
# Define function
bmi_giraffe <- function(giraffe) bmi(giraffe$weight, giraffe$height)

# Apply to all
map(info_giraffe, bmi_giraffe)
```

---
background-image: url(/assets/images/background_challenge_2.png)
class: left, middle

# Challenge 2

Using the *map strategy*:

1. Get 10 random numbers from a normal distribution (`rnorm(n = 10)`) for each of the mean values and return it as a list of 4:
```r
means <- c(-10, 0, 10, 20)
```
2. Read all sensor data as a list of data.frames by applying function `read_sensor_data` defined in script [`20200630_challenges.R`](https://github.com/inbo/coding-club/blob/master/src/20200630/20200630_challenges.R#L44-L52)

3. Get maximum and minimum datetime of all data.frames as a data.frame. Hint: write a function containing
```r
summarize(df, min_datetime = min(datetime), max_datetime = max(datetime))
```

4. Calculate 10 random numbers from a normal distribution (`rnorm(n = 10)`) for each of the mean and standard deviations below and return it as a list of 4. Hint: [mapping-over-multiple-arguments](https://r4ds.had.co.nz/iteration.html#mapping-over-multiple-arguments)
```r
means <- c(-10, 0, 10, 20)
st_dev <- c(1, 3, 2, 1.5)
```

---
class: left, middle

# `purrr` unleashed

Are you still hesitating about `purrr` potential? Well, check what `purrr` can do for your modelling work in the second page of the cheat sheet.

<center>![:scale 50%](/assets/images/20200630/20200630_second_page_cheatsheet_purrr.png)</center>

1. Group your data
2. Nest it to create nested data.frames (column containing data.frames!)
3. Work with it (apply modelling as in cheat sheet or whatever else you have in mind)

---
background-image: url(/assets/images/background_challenge_3.png)
class: left, middle

# Challenge 3

Let's play with [penguins](https://github.com/allisonhorst/palmerpenguins#palmerpenguins-) data! Group `penguins` data.frame by species and nest it as shown below:

```r
penguins %>%
 group_by(species) %>%
 nest()
# A tibble: 3 x 2
# Groups: species [3]
 species data
 <fct> <list>
1 Adelie <tibble [152 x 6]>
2 Gentoo <tibble [124 x 6]>
3 Chinstrap <tibble [68 x 6]>
```

Add a column called  `plot` containing ggplot objects based on the geometry and aesthetics below:

```r
geom_point(aes(x = bill_length_mm ,
               y = bill_depth_mm,
               colour = sex))
```

<center>![:scale 10%](/assets/images/20200630/20200630_palmer_penguins_data.png)</center>

---
class: left, middle

# Bonus challenge

1. Save the plots created in challenge 3 using species name and extension `png`. Hint: https://r4ds.had.co.nz/iteration.html#walk
2. purrr is more than the `map` family. To join the data.frames `species_1`, `species_2` and `species_3` defined in the script [20200630/20200630_challenges.R](https://github.com/inbo/coding-club/blob/master/src/20200630/20200630_challenges.R#L94-L113) you could apply `full_join()` twice:
```r
full_join(species_1, species_2, by = "species") %>%
  full_join(species_3, by = "species")
```
But this is definitely not DRY and prone to errors! Try to do it DRY by using lists and purrr. Hint: First page of the purrr cheat sheet.

---
class: left, middle

## Resources

- Beautiful webinar (pdf slides are downloadable) from Charlotte Wickham: [part 1](https://channel9.msdn.com/Events/useR-international-R-User-conferences/useR-International-R-User-2017-Conference/Solving-iteration-problems-with-purrr) and [part 2](https://channel9.msdn.com/Events/useR-international-R-User-conferences/useR-International-R-User-2017-Conference/Solving-iteration-problems-with-purrr-II). Introduction to `map()` and the map strategy material come from her.
- Well-written and complete tutorial: https://r4ds.had.co.nz/iteration.html#for-loops-vs.functionals (from 21.4 up to 21.9)
- Challenge solutions: [20200630/20200630_challenges_solutions.R](https://github.com/inbo/coding-club/blob/master/src/20200630/20200630_challenges_solutions.R)
- Video recording of the coding club session is available on [INBO's vimeo channel](https://vimeo.com/434253382)

---
class: center, middle

![:scale 30%](/assets/images/coding_club_logo_1.png)

Room: Herman Teirlinck - 01.21 - Jeanne Brabants 
Date: __25/08/2020__, van 10:00 tot 12:00 
Subject: debugging R
(registration announced via DG_useR@inbo.be)