useR to programmeR

👋 & Functions 1

Emma Rand and Ian Lyttle
WiFi: On your badges

👋 Welcome

Introductions

This is a one-day, hands-on workshop for those who have embraced the tidyverse and want to improve their R programming skills and, especially, reduce the amount of duplication in their code.

• do you have experience equivalent to an introductory data science course using tidyverse?
• are you comfortable with the Whole game chapter of R for Data Science (2nd Edition) by by Hadley Wickham, Mine Çetinkaya-Rundel, and Garrett Grolemund.

Material

• https://pos.it/r-programming-conf25

The team

Emma Rand

🦋@3mmarand.bsky.social

🐘@3mma@mastodon.social

Garrett Grolemund

Ian Lyttle

🦋@ijlyttle.bsky.social

🐘@ijlyttle@vis.social

 

Standing on the shoulders of

• R for Data Science (2e) Wickham, Çetinkaya-Rundel, and Grolemund (2023)

• The tidyverse style guide Wickham (n.d.)

• Programming with dplyr vignette Wickham et al. (2022)

WiFi

On your badges!

Introductions

To each other! With help from Yorkshire!

Ingredients: Sugar, Glucose syrup, Cocoa mass, Vegetable fats (Palm, Rapeseed, Sunflower, Coconut,Mango kernel/ Sal/ Shea), Sweetened condensed skimmed milk (Skimmed milk, Sugar), Cocoa butter, Dried whole milk, Glucose-fructose syrup, Coconut, Lactose and proteins from whey (from Milk), Whey powder (from Milk), Hazelnuts, Skimmed milk powder, Butter (from Milk), Emulsifiers (Sunflower lecithin, E471), Flavourings, Butterfat (from Milk), Fat-reduced cocoa powder, Salt, Lactic acid.

Ingredients: Sugar, glucose syrup, acid: citric acid; flavouring, colours: E102, E110, E122, E124, E133, E142.

Ingredients: Treacle, glucose syrup, maize starch, invert sugar syrup, wheat flour (wheat flour, calcium carbonate, iron, niacin, thiamin), liquorice extract, modified potato starch, vegetable oils (palm, sunflower, coconut), flavouring, glazing agent (carnauba wax).

Code of Conduct

Code of Conduct. Please Review

• 💙 Treat everyone with respect
• 🧡 Everyone should feel welcome and safe
• Red lanyard = ❌📷

Reporting:

• 🗣️ any posit::conf staff member (t-shirt) or Info desk
• 📧 codeofconduct@posit.com

Housekeeping

• Gender-neutral bathrooms:

  • LL1 - Off the back of the Centennial Foyer
  • LL2 - Next to Chicago A

• Meditation/prayer rooms: LL2 - Chicago A

• Lactation room: LL2 - Chicago B

🙏 to

• Garrett

• colleagues, friends and learners at Schneider Electric, University of York and RForwards!

• Posit team and especially Mine Çetinkaya-Rundel

• Ian!

• Experience 🍱 🥗 🌮 🍴 🕐

Prerequisites

We built this course using the most-recent versions of R (4.5) and RStudio (2025.05). However, things should work with at least R 4.2 and RStudio 2023.03. You will need packages:

• {tidyverse}
• {palmerpenguins}
• {devtools}
• {here}

🎬 Detailed instructions for installing these were covered in Prerequisites

Schedule

Time	Activity
09:00 - 10:30	Functions 1 Introduction, vector and dataframe functions, embracing
10:30 - 11:00	☕ Coffee break
11:00 - 12:30	Functions 2 Plot functions, style and side effects
12:30 - 13:30	🍱 🥗 🌮 🍴 Lunch break
13:30 - 15:00	Iteration 1 Introduction and modifying multiple columns
15:00 - 15:30	☕ Coffee break
15:30 - 17:00	Iteration 2 Reading and writing multiple files

How we will work

• stickies

  • 🟩 I’m all good, I’m done

  • 🎟️ I could do with some help

• Discord

• no stupid questions

• 🎬 Action!

Learning Objectives

At the end of this section you will be able to:

• explain the rationale for writing functions
• write vector functions
  • that take one or more vectors as input and output a vector
  • that take one or more vectors as input and output a single value
• specify defaults for function argument
• write functions that take dataframes as input and output a dataframe
• using embracing to allow data masking and tidy selection within functions

Set up

Project

https://github.com/posit-conf-2025/r-programming-exercises

🎬 Create a Project:

usethis::use_course("posit-conf-2025/r-programming-exercises")

> usethis::use_course("posit-conf-2025/r-programming-exercises")
✔ Downloading from 'https://github.com/posit-conf-2025/r-programming-exercises/zipball/HEAD'
Downloaded: 0.26 MB  
✔ Download stored in 'C:/Users/er13/OneDrive - University of York/Desktop/Desktop/posit-conf-2025-r-programming-exercises-978baff.zip'
✔ Unpacking ZIP file into 'posit-conf-2025-r-programming-exercises-978baff/' (45 files extracted)
Shall we delete the ZIP file ('posit-conf-2025-r-programming-exercises-978baff.zip')?

1: Not now
2: Yeah
3: Nope

🎬 Choose the option that means yes!

✔ Deleting 'posit-conf-2025-r-programming-exercises-978baff.zip'
✔ Opening project in RStudio

RStudio will restart

Create a .R

usethis::use_r("functions-01")

Packages

🎬 Load packages:

library(tidyverse)
library(palmerpenguins)

── Attaching core tidyverse packages ──────────────────────────────────────────────────────────── tidyverse 2.0.0 ──
✔ dplyr     1.1.2     ✔ readr     2.1.4
✔ forcats   1.0.0     ✔ stringr   1.5.0
✔ ggplot2   3.4.2     ✔ tibble    3.2.1
✔ lubridate 1.9.2     ✔ tidyr     1.3.0
✔ purrr     1.0.1     ── Conflicts ────────────────────────────────────────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag()    masks stats::lag()
ℹ Use the conflicted package to force all conflicts to become errors'

Using and loading penguins

🎬 Ensure we use palmerpenguins package over inbuilt:

conflicted::conflicts_prefer(palmerpenguins::penguins)

🎬 Load penguins data set

data(penguins)
glimpse(penguins)

Rows: 344
Columns: 8
$ species           <fct> Adelie, Adelie, Adelie, Adelie, Adelie, Adelie, Adel…
$ island            <fct> Torgersen, Torgersen, Torgersen, Torgersen, Torgerse…
$ bill_length_mm    <dbl> 39.1, 39.5, 40.3, NA, 36.7, 39.3, 38.9, 39.2, 34.1, …
$ bill_depth_mm     <dbl> 18.7, 17.4, 18.0, NA, 19.3, 20.6, 17.8, 19.6, 18.1, …
$ flipper_length_mm <int> 181, 186, 195, NA, 193, 190, 181, 195, 193, 190, 186…
$ body_mass_g       <int> 3750, 3800, 3250, NA, 3450, 3650, 3625, 4675, 3475, …
$ sex               <fct> male, female, female, NA, female, male, female, male…
$ year              <int> 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007…

Why write functions?

Rationale

• impact from code: reach and clarity
• efficiency: update code in one place, decrease error rate, improve ability to reuse

Example

We have several measurements:

• bill_length_mm
• bill_depth_mm
• flipper_length_mm
• body_mass_g

These are on very different scales

Example

• difficult to plot on same axis or determine what value is large for that variable

• A common solution is to apply a \(z\) score transformation to each variable.

• Normalises the values to have a mean of 0 and a standard deviation of 1

\[z = \frac{x - \bar{x}}{s.d.}\]

Apply transformation

We can apply the same transformation to each variable:

penguins <- penguins |>
  mutate(
    z_bill_length_mm = (bill_length_mm - mean(bill_length_mm, na.rm = TRUE)) / sd(bill_length_mm, na.rm = TRUE),
    z_bill_depth_mm = (bill_depth_mm - mean(bill_depth_mm, na.rm = TRUE)) / sd(bill_depth_mm, na.rm = TRUE),
    z_flipper_length_mm = (flipper_length_mm - mean(flipper_length_mm, na.rm = TRUE)) / sd(flipper_length_mm, na.rm = TRUE),
    z_body_mass_g = (body_mass_g - mean(body_mass_g, na.rm = TRUE)) / sd(body_mass_g, na.rm = TRUE)
  )

Long, unclear

(bill_length_mm - mean(bill_length_mm, na.rm = TRUE)) / sd(bill_length_mm, na.rm = TRUE)

• Quite a lot of code
• Difficult to determine what the transformation is

How to shorten and make more clear?

Coping and pasting

• is error prone

How to make fewer mistakes?

Writing a function:

• can be named to make transformation transparent
• will make code shorter
• can be reused

🔑️ You may think you have to write complex functions - you don’t! Start with the simple things.

Types of function

Types of function

We will cover two types of function

1. vector functions: one or more vectors as input, one vector as output

2. data frame functions: df as input and df as output

Types of function

We will cover two types of function

1. vector functions: one of more vectors as input, one vector as output

   1. output length is same as input length. These functions will work well in mutate() and filter().

   2. summary functions: input is vector, output is a single value

2. data frame functions: df as input and df as output

Vector functions

Output same length as input

• output same length as input
• work well in mutate()
• appropriate for the z-transformation example

General

To turn your code into a function you need:

• a name
• the arguments - which represent the bits that vary
• the code body for the function

name <- function(arguments) {
  code body
}

Function name

Use a verb - The tidyverse style guide (Wickham, n.d.) but good advice regardless

Difficulty in naming? Should this be two or three functions?

What should we call the function we write to do a \(z\) score transformation?

Arguments

• the input vector

• additional arguments

Naming conventions

• x for the vector input

name <- function(x) {
  body does things with x
}

Example

\[z = \frac{x - \bar{x}}{s.d.}\]

penguins <- penguins |>
  mutate(
    z_bill_length_mm = (bill_length_mm - mean(bill_length_mm, na.rm = TRUE)) / sd(bill_length_mm, na.rm = TRUE),
    z_bill_depth_mm = (bill_depth_mm - mean(bill_depth_mm, na.rm = TRUE)) / sd(bill_depth_mm, na.rm = TRUE),
    z_flipper_length_mm = (flipper_length_mm - mean(flipper_length_mm, na.rm = TRUE)) / sd(flipper_length_mm, na.rm = TRUE),
    z_body_mass_g = (body_mass_g - mean(body_mass_g, na.rm = TRUE)) / sd(body_mass_g, na.rm = TRUE)
  )

Example

Identify the arguments: the things that vary across calls

(bill_length_mm    - mean(bill_length_mm,    na.rm = TRUE)) / sd(bill_length_mm,    na.rm = TRUE)
(bill_depth_mm     - mean(bill_depth_mm,     na.rm = TRUE)) / sd(bill_depth_mm,     na.rm = TRUE)
(flipper_length_mm - mean(flipper_length_mm, na.rm = TRUE)) / sd(flipper_length_mm, na.rm = TRUE)
(body_mass_g       - mean(body_mass_g,       na.rm = TRUE)) / sd(body_mass_g,       na.rm = TRUE)

(🟧 - mean(🟧, na.rm = TRUE)) / sd(🟧, na.rm = TRUE)
(🟧 - mean(🟧, na.rm = TRUE)) / sd(🟧, na.rm = TRUE)
(🟧 - mean(🟧, na.rm = TRUE)) / sd(🟧, na.rm = TRUE)
(🟧 - mean(🟧, na.rm = TRUE)) / sd(🟧, na.rm = TRUE)

🟧 is x

Example

Put into the template

name <- function(x) {
  body does things with x
}

to_z <- function(x) {
  (x - mean(x, na.rm = TRUE)) / sd(x, na.rm = TRUE)
}

Apply

Rewrite the call to mutate() as:

penguins <- penguins |>
  mutate(
    z_bill_length_mm = to_z(bill_length_mm),
    z_bill_depth_mm = to_z(bill_depth_mm),
    z_flipper_length_mm = to_z(flipper_length_mm),
    z_body_mass_g = to_z(body_mass_g)
  )

Much shorter, much more clear.

A modification

mean() has a trim argument: mean(x, trim = 0, na.rm = FALSE, ...)

the fraction (0 to 0.5) of observations to be trimmed from each end of x before the mean is computed.

Suppose we want to specify the middle proportion left rather than the proportion trimmed from each end.

A modification

• A value of 0.1 for trim trims 0.1 from each end leaving 0.8 in the middle

• trim = (1 - middle)/2

Trim is the proportion trimmed off each end; middle is what’s left

Add an argument

to_z <- function(x, middle) {
  trim = (1 - middle)/2
  (x - mean(x, na.rm = TRUE, trim = trim)) / sd(x, na.rm = TRUE)
}

Try it out

to_z(penguins$bill_length_mm, middle = 0.2)

  [1] -0.92838057 -0.85511491 -0.70858359          NA -1.36797452 -0.89174774
  [7] -0.96501340 -0.91006415 -1.84420131 -0.39720454 -1.16649396 -1.16649396
 [13] -0.56205227 -1.01996264 -1.75261923 -1.38629094 -1.00164623 -0.30562246
 [19] -1.78925206  0.33545205 -1.16649396 -1.18481038 -1.51450584 -1.09322830
 [25] -0.98332981 -1.62440433 -0.65363435 -0.67195076 -1.14817755 -0.67195076
 [31] -0.85511491 -1.27639245 -0.85511491 -0.59868510 -1.42292377 -0.91006415
 [37] -0.98332981 -0.36057171 -1.20312679 -0.80016566 -1.40460735 -0.61700152
 [43] -1.49618943 -0.01255983 -1.31302528 -0.83679849 -0.56205227 -1.22144320
 [49] -1.49618943 -0.34225529 -0.83679849 -0.74521642 -1.67935358 -0.39720454
 [55] -1.77093565 -0.50710303 -0.94669698 -0.65363435 -1.40460735 -1.20312679
 [61] -1.55113867 -0.52541944 -1.20312679 -0.56205227 -1.42292377 -0.47047020
 [67] -1.58777150 -0.56205227 -1.51450584 -0.43383737 -1.95409980 -0.81848208
 [73] -0.83679849  0.29881922 -1.58777150 -0.25067322 -0.59868510 -1.27639245
 [79] -1.45955660 -0.37888812 -1.75261923 -0.23235681 -1.36797452 -1.66103716
 [85] -1.25807603 -0.52541944 -1.44124018 -1.33134169 -1.07491189 -0.96501340
 [91] -1.55113867 -0.56205227 -1.86251772 -0.83679849 -1.45955660 -0.61700152
 [97] -1.11154472 -0.70858359 -2.02736546 -0.17740756 -1.67935358 -0.58036869
[103] -1.18481038 -1.16649396 -1.14817755 -0.81848208 -1.01996264 -1.09322830
[109] -1.11154472 -0.17740756 -1.11154472  0.26218639 -0.81848208 -0.36057171
[115] -0.83679849 -0.26898963 -1.01996264 -1.25807603 -1.55113867 -0.56205227
[121] -1.45955660 -1.18481038 -0.72690000 -0.50710303 -1.64272075 -0.65363435
[127] -0.98332981 -0.48878661 -0.94669698 -0.01255983 -1.03827906 -0.19572398
[133] -1.34965811 -1.22144320 -1.11154472 -0.56205227 -1.56945509 -0.72690000
[139] -1.31302528 -0.81848208 -0.72690000 -0.65363435 -2.21052960 -0.63531793
[145] -1.25807603 -0.94669698 -0.91006415 -1.38629094 -1.49618943 -1.16649396
[151] -1.49618943 -0.48878661  0.35376847  1.06810865  0.82999525  1.06810865
[157]  0.62851469  0.42703413  0.22555357  0.46366696 -0.15909115  0.48198337
[163] -0.59868510  0.88494450  0.24386998  0.77504601  0.29881922  0.93989374
[169] -0.39720454  0.92157733  0.37208488  0.82999525  1.10474148  0.17060432
[175]  0.42703413  0.39040130 -0.23235681  0.35376847  0.06070583  0.66514752
[181]  0.73841318  1.06810865  0.57356545 -0.25067322  0.17060432  2.82648447
[187]  0.90326091  0.77504601 -0.28730605  0.04238942 -0.03087624  0.82999525
[193] -0.26898963  0.99484299  0.20723715  0.99484299  1.15969072 -0.10414190
[199]  0.24386998  1.15969072  0.13397149  0.18892074  0.44535054  0.79336242
[205]  0.17060432  1.08642506  0.42703413  0.15228791 -0.06750907  0.24386998
[211] -0.17740756  1.14137431  0.20723715  0.37208488  0.28050281  1.85571448
[217]  0.29881922  1.03147582  0.37208488  0.97652657 -0.12245832  1.19632355
[223]  0.64683111  0.40871771  0.73841318  0.42703413  0.40871771  0.81167884
[229]  0.61019828  1.26958921  0.18892074  0.18892074  0.90326091  1.52601902
[235]  0.59188186  1.06810865  0.13397149  1.21463997 -0.14077473  1.30622204
[241]  0.61019828  1.45275336  0.61019828  1.47106977  0.24386998  0.97652657
[247]  0.06070583  1.21463997  0.95821016  0.50029979  0.77504601  1.26958921
[253]  0.79336242  2.14877712  0.55524903  0.90326091  0.57356545  0.48198337
[259] -0.45215378  1.69086675 -0.15909115  0.72009677  1.15969072  1.03147582
[265] -0.12245832  1.34285487  0.37208488  2.00224580  0.06070583  0.84831167
[271]  0.55524903          NA  0.48198337  1.14137431  0.18892074  1.04979223
[277]  0.42703413  1.06810865  1.30622204  0.22555357  1.56265185  0.18892074
[283]  0.35376847  1.30622204  0.33545205  1.30622204  0.44535054  1.37948770
[289]  0.51861620  1.43443694  0.31713564  1.15969072  1.12305789  2.53342183
[295]  0.40871771  0.92157733 -0.32393888  0.79336242 -0.17740756  1.17800714
[301]  0.46366696  1.43443694  1.15969072  0.97652657  0.40871771  1.58096826
[307] -0.59868510  1.83739807 -0.30562246  1.25127279  1.01315940  0.61019828
[313]  0.62851469  1.43443694  0.50029979  1.70918316  0.88494450  0.37208488
[319]  1.23295638  0.24386998  1.23295638  1.21463997  1.08642506  0.88494450
[325]  1.34285487  1.03147582  0.72009677  1.32453845  0.28050281  1.19632355
[331] -0.30562246  1.47106977  0.18892074  0.93989374  1.10474148  0.26218639
[337]  1.41612053  0.48198337  0.28050281  2.13046071 -0.12245832  0.99484299
[343]  1.21463997  1.10474148

But what if we forget?

to_z(penguins$bill_length_mm)

Error in to_z(penguins$bill_length_mm): argument "middle" is missing, with no default

Give a default

Give defaults whenever possible:

to_z <- function(x, middle = 1) {
  trim = (1 - middle)/2
  (x - mean(x, na.rm = TRUE, trim = trim)) / sd(x, na.rm = TRUE)
}

Try it out

to_z(penguins$bill_length_mm)

  [1] -0.88320467 -0.80993901 -0.66340769          NA -1.32279862 -0.84657184
  [7] -0.91983750 -0.86488825 -1.79902541 -0.35202864 -1.12131806 -1.12131806
 [13] -0.51687637 -0.97478674 -1.70744334 -1.34111504 -0.95647033 -0.26044656
 [19] -1.74407616  0.38062795 -1.12131806 -1.13963448 -1.46932994 -1.04805240
 [25] -0.93815391 -1.57922843 -0.60845845 -0.62677486 -1.10300165 -0.62677486
 [31] -0.80993901 -1.23121655 -0.80993901 -0.55350920 -1.37774787 -0.86488825
 [37] -0.93815391 -0.31539581 -1.15795089 -0.75498976 -1.35943145 -0.57182562
 [43] -1.45101353  0.03261607 -1.26784938 -0.79162259 -0.51687637 -1.17626731
 [49] -1.45101353 -0.29707939 -0.79162259 -0.70004052 -1.63417768 -0.35202864
 [55] -1.72575975 -0.46192713 -0.90152108 -0.60845845 -1.35943145 -1.15795089
 [61] -1.50596277 -0.48024354 -1.15795089 -0.51687637 -1.37774787 -0.42529430
 [67] -1.54259560 -0.51687637 -1.46932994 -0.38866147 -1.90892390 -0.77330618
 [73] -0.79162259  0.34399512 -1.54259560 -0.20549732 -0.55350920 -1.23121655
 [79] -1.41438070 -0.33371222 -1.70744334 -0.18718091 -1.32279862 -1.61586126
 [85] -1.21290014 -0.48024354 -1.39606428 -1.28616579 -1.02973599 -0.91983750
 [91] -1.50596277 -0.51687637 -1.81734182 -0.79162259 -1.41438070 -0.57182562
 [97] -1.06636882 -0.66340769 -1.98218956 -0.13223166 -1.63417768 -0.53519279
[103] -1.13963448 -1.12131806 -1.10300165 -0.77330618 -0.97478674 -1.04805240
[109] -1.06636882 -0.13223166 -1.06636882  0.30736229 -0.77330618 -0.31539581
[115] -0.79162259 -0.22381374 -0.97478674 -1.21290014 -1.50596277 -0.51687637
[121] -1.41438070 -1.13963448 -0.68172411 -0.46192713 -1.59754485 -0.60845845
[127] -0.93815391 -0.44361071 -0.90152108  0.03261607 -0.99310316 -0.15054808
[133] -1.30448221 -1.17626731 -1.06636882 -0.51687637 -1.52427919 -0.68172411
[139] -1.26784938 -0.77330618 -0.68172411 -0.60845845 -2.16535371 -0.59014203
[145] -1.21290014 -0.90152108 -0.86488825 -1.34111504 -1.45101353 -1.12131806
[151] -1.45101353 -0.44361071  0.39894437  1.11328455  0.87517115  1.11328455
[157]  0.67369059  0.47221003  0.27072946  0.50884286 -0.11391525  0.52715927
[163] -0.55350920  0.93012040  0.28904588  0.82022191  0.34399512  0.98506964
[169] -0.35202864  0.96675323  0.41726078  0.87517115  1.14991738  0.21578022
[175]  0.47221003  0.43557720 -0.18718091  0.39894437  0.10588173  0.71032342
[181]  0.78358908  1.11328455  0.61874135 -0.20549732  0.21578022  2.87166037
[187]  0.94843681  0.82022191 -0.24213015  0.08756532  0.01429966  0.87517115
[193] -0.22381374  1.04001889  0.25241305  1.04001889  1.20486662 -0.05896600
[199]  0.28904588  1.20486662  0.17914739  0.23409663  0.49052644  0.83853832
[205]  0.21578022  1.13160096  0.47221003  0.19746381 -0.02233317  0.28904588
[211] -0.13223166  1.18655021  0.25241305  0.41726078  0.32567871  1.90089038
[217]  0.34399512  1.07665172  0.41726078  1.02170247 -0.07728242  1.24149945
[223]  0.69200701  0.45389361  0.78358908  0.47221003  0.45389361  0.85685474
[229]  0.65537418  1.31476511  0.23409663  0.23409663  0.94843681  1.57119492
[235]  0.63705776  1.11328455  0.17914739  1.25981586 -0.09559883  1.35139794
[241]  0.65537418  1.49792926  0.65537418  1.51624567  0.28904588  1.02170247
[247]  0.10588173  1.25981586  1.00338606  0.54547569  0.82022191  1.31476511
[253]  0.83853832  2.19395302  0.60042493  0.94843681  0.61874135  0.52715927
[259] -0.40697788  1.73604265 -0.11391525  0.76527266  1.20486662  1.07665172
[265] -0.07728242  1.38803077  0.41726078  2.04742170  0.10588173  0.89348757
[271]  0.60042493          NA  0.52715927  1.18655021  0.23409663  1.09496813
[277]  0.47221003  1.11328455  1.35139794  0.27072946  1.60782775  0.23409663
[283]  0.39894437  1.35139794  0.38062795  1.35139794  0.49052644  1.42466360
[289]  0.56379210  1.47961284  0.36231154  1.20486662  1.16823379  2.57859773
[295]  0.45389361  0.96675323 -0.27876298  0.83853832 -0.13223166  1.22318303
[301]  0.50884286  1.47961284  1.20486662  1.02170247  0.45389361  1.62614416
[307] -0.55350920  1.88257397 -0.26044656  1.29644869  1.05833530  0.65537418
[313]  0.67369059  1.47961284  0.54547569  1.75435906  0.93012040  0.41726078
[319]  1.27813228  0.28904588  1.27813228  1.25981586  1.13160096  0.93012040
[325]  1.38803077  1.07665172  0.76527266  1.36971435  0.32567871  1.24149945
[331] -0.26044656  1.51624567  0.23409663  0.98506964  1.14991738  0.30736229
[337]  1.46129643  0.52715927  0.32567871  2.17563660 -0.07728242  1.04001889
[343]  1.25981586  1.14991738

Your turn

The Box-Cox transformation is a statistical technique that converts non-normally distributed data to an approximately normal distribution by applying a power-transformation based on a parameter lambda (\(\lambda\)).

The transformation on a variable \(x\) is:

\[bc = \frac{x^{\lambda} - 1}{\lambda} \]

You don’t need to understand this!

Your turn: Box-Cox

🎬 Your task is to write a function that performs the Box-Cox power transformation

🎬 First, create a vector of values for testing the function:

vals <- rexp(10000, 10) 

Effect of Box-Cox transformation

Distribution of values before (left) and after (right) a Box-Cox transformation.

Your turn: Box-Cox

1. Set the default \(\lambda = 1\)

2. Still have time? In fact, the transformation only applies when \(\lambda \ne 0\). Check if \(\lambda = 0\) and if it does, set it to 0.0001 and print “Lambda cannot be 0, setting to 0.0001”

3. Still have time? In fact, the transformation is defined as \(log(x)\) when \(\lambda = 0\). Amend the function so it performs the appropriate transformation depending on the value of \(\lambda\).

\[ bc = \begin{cases} \frac{x^{\lambda} - 1}{\lambda} & \text{for }\lambda \ne 0\\ log(x) & \text{for }\lambda = 0 \end{cases} \]

1. A solution

to_box_cox <- function(x, lambda = 1) {
  (x^lambda - 1) / lambda
}

Test:

to_box_cox(vals, 0.3) |> 
  hist()

2. A solution

to_box_cox <- function(x, lambda = 1) {
  if (lambda == 0) {
    lambda <- 0.0001
    message("Lambda cannot be 0, setting to 0.0001")
  }
  (x^lambda - 1) / lambda
}

Test:

to_box_cox(vals, 0) |> 
  hist()

3. A solution

to_box_cox <- function(x, lambda = 1) {
  if (lambda == 0) {
    log(x)
  }
  else {
    (x^lambda - 1) / lambda
  }
}

Test:

to_box_cox(vals, 0) |> 
  hist()

Types of function

We will cover two types of function

1. vector functions: one or more vectors as input, one vector as output

   1. ✔️ output same length as input.

   ii. ➡️ summary functions: input is vector, output is a single value

2. data frame functions: df as input and df as output

Summary functions

• input is vector
• output is a single value
• could be used in summarise()

Example

Write a function to compute the standard error of a sample.

\[s.e. = \frac{s.d.}{\sqrt{n}}\]

Example

sd_error <- function(x){
  sd(x, na.rm = TRUE) / sqrt(sum(!is.na(x)))
}

Note: sum(TRUE) = 1 and sum(FALSE) = 0 Thus,sum(!is.na(x)) gives you the number of TRUE (i.e., the number of non-NA values) and is a bit shorter than length(x[!is.na(x)])

Try it out

🎬 Call the function on penguins$bill_length_mm

sd_error(penguins$bill_length_mm)

[1] 0.2952205

Or in a pipeline

penguins |> 
  summarise(se = sd_error(bill_length_mm))

# A tibble: 1 × 1
     se
  <dbl>
1 0.295

Your turn

🎬 Write a function to compute the sums of squares (sum of the squared deviations from the mean). This can be calculated with either:

\[SS(x) = \sum{(x - \bar{x})^2}\]

or

\[SS(x) = s^2 * (n-1)\]

A solution - one option

sum_sq <- function(x){
 sum((x - mean(x, na.rm = TRUE))^2, na.rm = TRUE)
}

🎬 Try it out

sum_sq(penguins$bill_length_mm)

[1] 10164.21

Types of function

We will cover two types of function

1. vector functions: one or more vectors as input, one vector as output

   1. ✔️ output same length as input.

   2. ✔️ summary functions: input is vector, output is a single value

2. ➡️ data frame functions: df as input and df as output

Dataframe functions

Dataframe functions

Dataframe as input and Dataframe as output

For example, we might summarise one of our columns like this:

penguins |> 
  summarise(mean = mean(bill_length_mm, na.rm = TRUE),
            n = sum(!is.na(bill_length_mm)),
            sd = sd(bill_length_mm, na.rm = TRUE),
            se = sd_error(bill_length_mm))

# A tibble: 1 × 4
   mean     n    sd    se
  <dbl> <int> <dbl> <dbl>
1  43.9   342  5.46 0.295

Output is a dataframe

Dataframe functions

and summarise several dataframes in the same way

Good candidate for a function to avoid repetitive code: my_summary()

Define my_summary() function

my_summary <- function(df, column){
  df |> 
  summarise(mean = mean(column, na.rm = TRUE),
            n = sum(!is.na(column)),
            sd = sd(column, na.rm = TRUE),
            se = sd_error(column))
}

Use function

my_summary(penguins, bill_length_mm)

Error in `summarise()`:
ℹ In argument: `mean = mean(column, na.rm = TRUE)`.
Caused by error:
! object 'bill_length_mm' not found

😕

Tidy evaluation

tidyverse functions like dplyr::summarise() use “tidy evaluation” so you can refer to the names of variables inside dataframes. For example, you can use:

either

penguins |> summarise(mean = mean(bill_depth_mm))

Or

summarise(penguins, mean = mean(bill_depth_mm))

Tidy evaluation

This is instead of having to use the full dataframe name with $, e.g.

summarise(penguins, mean = mean(penguins$bill_depth_mm))

This is known as data-masking: the dataframe environment masks the user environment by giving priority to the dataframe.

Data masking is great….

and makes life easier when working interactively

But not so useful in functions

Because of data-masking, summarise() in my_summary() is looking for a column literally called column in the dataframe that has been passed in. It is not looking inside the variable column for the name of column you want to give it.

Read more: Programming with dplyr

Fix my_summary() function

The solution is to use embracing: { var }

my_summary <- function(df, column){
  df |> 
  summarise(mean = mean({{ column }}, na.rm = TRUE),
            n = sum(!is.na({{ column }})),
            sd = sd({{ column }}, na.rm = TRUE),
            se = sd_error({{ column }}),
            .groups = "drop")
}

• look inside column variable
• style with spaces
• .groups = "drop" to avoid message and leave the data in an ungrouped state

Use function

my_summary(penguins, bill_length_mm)

# A tibble: 1 × 4
   mean     n    sd    se
  <dbl> <int> <dbl> <dbl>
1  43.9   342  5.46 0.295

🎉

When to embrace?

When tidy evaluation is used

Your turn

🎬 Write a new summary function which calculates the median, maximum and minimum values of a variable. Incorporate an argument to allow the summary to be performed grouped by another variable.

A solution - 1

my_summary <- function(df, summary_var, group_var){
  df |> 
    group_by({{ group_var }}) |> 
  summarise(median = median({{summary_var  }}, na.rm = TRUE),
            minimum = min({{summary_var  }}, na.rm = TRUE),
            maximum = max({{summary_var  }}, na.rm = TRUE),
            .groups = "drop")
}

Your turn

🎬 Try it out

my_summary(penguins, bill_length_mm, species)

# A tibble: 3 × 4
  species   median minimum maximum
  <fct>      <dbl>   <dbl>   <dbl>
1 Adelie      38.8    32.1    46  
2 Chinstrap   49.6    40.9    58  
3 Gentoo      47.3    40.9    59.6

A solution - 2

Improvement: Have a default of NULL for the grouping variable

my_summary <- function(df, summary_var, group_var = NULL){
  df |> 
    group_by({{ group_var }}) |> 
    summarise(median = median({{summary_var  }}, na.rm = TRUE),
            minimum = min({{summary_var  }}, na.rm = TRUE),
            maximum = max({{summary_var  }}, na.rm = TRUE),
            .groups = "drop")
}

Your turn

🎬 Try it out

my_summary(penguins, bill_length_mm)

# A tibble: 1 × 3
  median minimum maximum
   <dbl>   <dbl>   <dbl>
1   44.4    32.1    59.6

Your turn

🎬 Try it out with more than one group

my_summary(penguins, bill_length_mm, c(species, island))

Error in `group_by()`:
ℹ In argument: `c(species, island)`.
Caused by error:
! `c(species, island)` must be size 344 or 1, not 688.

😕

A solution - 3

Use pick() which allows you to select a subset of columns inside a data masking function:

my_summary <- function(df, summary_var, group_var = NULL){
  df |> 
    group_by(pick({{ group_var }})) |> 
    summarise(median = median({{summary_var  }}, na.rm = TRUE),
            minimum = min({{summary_var  }}, na.rm = TRUE),
            maximum = max({{summary_var  }}, na.rm = TRUE),
            .groups = "drop")
}

🎬 Try it out with more than one group

my_summary(penguins, bill_length_mm, c(species, island))

# A tibble: 5 × 5
  species   island    median minimum maximum
  <fct>     <fct>      <dbl>   <dbl>   <dbl>
1 Adelie    Biscoe      38.7    34.5    45.6
2 Adelie    Dream       38.6    32.1    44.1
3 Adelie    Torgersen   38.9    33.5    46  
4 Chinstrap Dream       49.6    40.9    58  
5 Gentoo    Biscoe      47.3    40.9    59.6

Extras

• Short cuts:

  • put cursor on a function call and press F2 to find its definition
  • Ctrl+. opens section/file search

Summary ☕

• Writing functions can make you more efficient and make your code more readable. This can be just for your benefit.

• Vector functions take one or more vectors as input; their output can be a vector (useful in mutate() and filter()) or a single value (useful in summarise())

• Dataframe functions take a dataframe as input and give a dataframe as output

• Give arguments a default where possible

• We use { var } embracing to manage data masking

• We use pick() to select more than one variable

References

Wickham, Hadley. n.d. Welcome | The Tidyverse Style Guide. https://style.tidyverse.org/index.html.

Wickham, Hadley, Mine Çetinkaya-Rundel, and Garrett Grolemund. 2023. R for Data Science. 2nd ed. https://r4ds.hadley.nz/.

Wickham, Hadley, Romain François, Lionel Henry, and Kirill Müller. 2022. “Dplyr: A Grammar of Data Manipulation.” https://CRAN.R-project.org/package=dplyr.