Will your code run again?
Tips for making code reproducible in R

Dax Kellie, Shandiya Balasubramaniam, Martin Westgate

@daxkellie.bsky.social

I acknowledge the Traditional Owners of the lands on which I live and work, the Dharawal, Wandi Wandian, and Gadigal people of the Yuin Nation, and pay my respects to Elders past and present. I recognise the significance of this environment to Traditional Owners, and acknowledge their continuing connection to Country.

Lots of researchers use code

…and only some code works correctly

Whether a code script reproduces results usually has a lower chance than a coin toss

Code reproducibility is hard

Self-trained coders
Packages constantly update to newer versions
The environment and organisms we study have lots of variation
Academic institutions & funding schemes incentivise speed & “novel” results over reproducibility & longevity
Researchers have lots to do

…but maybe it’s just hard to change…

Code availability requirements

Journals require code availability statements
Peer reviewers can require code is provided

Reproducibility is fundamental to science

Supporting reproducibility at the ALA

https://galah.ala.org.au/

https://labs.ala.org.au/

https://cleaning-data-r.ala.org.au/

Let’s make a code snippet reproducible

Our code snippet

library(tidyverse)
library(galah)
setwd("C:\Users\KEL329\OneDrive - CSIRO\Documents\ALA\Talks\ESA2024")

galah_config(email = "dax.kellie@csiro.au")
alaData <- galah_call()|>identify("perameles")|>filter(year == 2003) |>
  select(group="basic",cl22)|>
  atlas_occurrences()|>
  select(recordID,scientificName,decimalLongitude,decimalLatitude,eventDate,cl22) |>
  janitor::clean_names()|>rename(state=cl22) |>
  mutate(event_date=lubridate::ymd(event_date)) |>
  group_by(state) |>count() |>drop_na()
mmap <- ozmaps::ozmap_states |>
  sf::st_transform(crs=4326) |>
  left_join(alaData, join_by(NAME==state)) |>
  replace_na(list(n=0))

ggplot() + geom_sf(data=mmap,aes(fill=n),colour="grey60")+
  viridis::scale_fill_viridis(option="F",begin=0.2,direction=-1)+theme_void()+theme(legend.position="right")

Our code snippet

# Title: Map - number of bandicoot observations
# Author: Dax Kellie
# Date: 2024-11-16

setwd("C:\Users\KEL329\OneDrive - CSIRO\Documents\ALA\Talks\ESA2024")

# packages
library(tidyverse)
library(galah)
library(janitor)
library(sf)
library(ozmaps)

galah_config(email = "dax.kellie@csiro.au")

# download map of Australia
aus <- ozmap_states |>
  st_transform(crs = 4326) # fix projection

# download bandicoot records
bandicoots <- galah_call() |>
  identify("perameles") |>
  filter(year == 2003) |>
  select(group = "basic", cl22) |>
  atlas_occurrences() 

# filter data, rename column, fix date class
bandicoots |>
  select(recordID, scientificName, decimalLongitude, 
         decimalLatitude, eventDate, cl22) |>
  janitor::clean_names() |> 
  rename(state = cl22) |>
  mutate(
    event_date = lubridate::ymd(event_date)
    ) 

# counts by state/territory
state_counts <- 
  bandicoots_cleaned |>
  group_by(state) |> 
  count() |> 
  drop_na()

# join map with counts
aus_counts <- 
  aus |>
  left_join(state_counts, join_by(NAME == state)) |>
  replace_na(list(n = 0))

# Map
ggplot() + 
  geom_sf(data = aus_counts,
          aes(fill = n),
          colour = "grey60") +
  viridis::scale_fill_viridis(option = "F", 
                              begin = 0.2, 
                              direction = -1) + 
  theme_void() + 
  theme(legend.position = "right")

This code reads better, but it’s no more reproducible!
Code reproducibilty depends on a reproducible work environment

It’s all about the setup
aka a reproducible project environment

6 simple steps
to make your R code run again

1. Create an R Project

R projects use .Rproj files to tell R where your project’s top folder directory is

Works nicely with the {here} package (for safe file paths)!

2. Use GitHub

An online platform for storing project repositories

2. Use GitHub

GitHub is useful for reasons other than collaborative code writing, too!

Front-facing README files improve project documentation
Easy to share projects with others
Find old code. Copy/Paste it back in

2. Use GitHub

To get setup:

Create a GitHub account
Introduce yourself to Git & R:
Happy Git and GitHub for the useR

Tip: Setup GitHub with {usethis}

use_git() + use_github() can initialise & link a local directory to a GitHub repository and it’s fast

3. Organised folder structure

Straightforward folder structure
This might require ongoing maintenance

4. Readable file names

Jenny Bryan’s file name holy trinity

machine readable
human readable
plays well with default ordering

4. Readable file names

Bad:

dat2024_bsrFinalDK-new.csv
script.R

Good:

2024-16-11_bandicoots.csv
map_counts-by-state.R

5. Record packages & versions

In order of comprehensiveness/ease:

{renv}
- init(), snapshot(), restore()
{groundhog}
- groundhog.library(pkg-name, date)
sessionInfo()
- sessionInfo() |> report::report()

6. Back up your data

Locally & online

Zenodo, Open Science Framework
- Allow embargoes or restricted access (in case of private data)

Generate a DOI for your data
- DOIs persist even if urls change

6. Back up your data

Generate a DOI for your data

library(galah)
galah_config(email = "dax.kellie@csiro.au", verbose = FALSE) # ALA email

# download data
bandicoots <- galah_call() |>
  identify("perameles") |>
  filter(year == 2004) |>
  atlas_occurrences(mint_doi = TRUE) # add a data DOI

attributes(bandicoots)$doi # see DOI

[1] "https://doi.org/10.26197/ala.78e21acd-7516-4fb2-91fe-9b86f5fcd83b"

# retrieve data
galah_call() |>
  filter(doi == "https://doi.org/10.26197/ala.974f0355-5ad8-439e-91c1-709a49df1086") |>
  atlas_occurrences()

Let’s return to our code snippet

Our code snippet

# Title: Map - number of bandicoot observations
# Author: Dax Kellie
# Date: 2024-11-16

setwd("C:\Users\KEL329\OneDrive - CSIRO\Documents\ALA\Talks\ESA2024")

# packages
library(tidyverse)
library(galah)
library(janitor)
library(sf)
library(ozmaps)

galah_config(email = "dax.kellie@csiro.au")

# download map of Australia
aus <- ozmap_states |>
  st_transform(crs = 4326) # fix projection

# download bandicoot records
bandicoots <- galah_call() |>
  identify("perameles") |>
  filter(year == 2003) |>
  select(group = "basic", cl22) |>
  atlas_occurrences() 

# filter data, rename column, fix date class
bandicoots |>
  select(recordID, scientificName, decimalLongitude, 
         decimalLatitude, eventDate, cl22) |>
  janitor::clean_names() |> 
  rename(state = cl22) |>
  mutate(
    event_date = lubridate::ymd(event_date)
    ) 

# counts by state/territory
state_counts <- 
  bandicoots_cleaned |>
  group_by(state) |> 
  count() |> 
  drop_na()

# join map with counts
aus_counts <- 
  aus |>
  left_join(state_counts, join_by(NAME == state)) |>
  replace_na(list(n = 0))

# Map
ggplot() + 
  geom_sf(data = aus_counts,
          aes(fill = n),
          colour = "grey60") +
  viridis::scale_fill_viridis(option = "F", 
                              begin = 0.2, 
                              direction = -1) + 
  theme_void() + 
  theme(legend.position = "right")

Our code snippet

# Title: Map - number of bandicoot observations
# Author: Dax Kellie
# Date: 2024-11-16



# packages
library(tidyverse)
library(galah)
library(janitor)
library(sf)
library(ozmaps)

galah_config(email = "dax.kellie@csiro.au")

# download map of Australia
aus <- ozmap_states |>
  st_transform(crs = 4326) # fix projection

# download bandicoot records
bandicoots <- galah_call() |>
  identify("perameles") |>
  filter(year == 2003) |>
  select(group = "basic", cl22) |>
  atlas_occurrences() 

# filter data, rename column, fix date class
bandicoots |>
  select(recordID, scientificName, decimalLongitude, 
         decimalLatitude, eventDate, cl22) |>
  janitor::clean_names() |> 
  rename(state = cl22) |>
  mutate(
    event_date = lubridate::ymd(event_date)
    ) 

# counts by state/territory
state_counts <- 
  bandicoots_cleaned |>
  group_by(state) |> 
  count() |> 
  drop_na()

# join map with counts
aus_counts <- 
  aus |>
  left_join(state_counts, join_by(NAME == state)) |>
  replace_na(list(n = 0))

# Map
ggplot() + 
  geom_sf(data = aus_counts,
          aes(fill = n),
          colour = "grey60") +
  viridis::scale_fill_viridis(option = "F", 
                              begin = 0.2, 
                              direction = -1) + 
  theme_void() + 
  theme(legend.position = "right")

Summary

Your code will run again!

Making code reproducibility depends on making a reproducible working environment

- R projects
- GitHub
- Organised folder
- Well-named files
- Document package versions
- Data stored locally & online

On reproducibility

When 174 analyst teams were asked to use 2 datasets to answer 2 ecology/evolution questions, results were all over the grid.

On reproducibility

Every decision a researcher makes affects the result and its interpretation.
To interpret scientific evidence, one must be able to reproduce & interrogate the analytic steps that led to a result.

Other useful resources

Setting up a reproducible data analysis project in R
Talk by Olivia Angelin-Bonnet 2024
Improving code reproducibility: Small steps with big impacts
Nature EcoEvo Blog by Dax Kellie & Martin Westgate
Implementing code review in the scientific workflow: Insights from ecology and evolutionary biology
Paper by Ed Ivimey Cook et al 2023
A how-to guide for code sharing in biology
Paper by Richard Abdill et al 2023

Thank you

Dax Kellie
Data Analyst & Science Lead
Science & Decision Support | ALA
: dax.kellie@csiro.au
: @daxkellie.bsky.social
: @daxkellie

Science & Decision Support team
Martin Westgate, Shandiya Balasubramaniam
Olivia Torresan, Juliet Seers, Amanda Buyan

These slides were made using Quarto & RStudio

Slides:

Will your code run again?Tips for making code reproducible in R

Lots of researchers use code

…but few researchers share code…

…and only some code works correctly

Code reproducibility is hard

Code availability requirements

Reproducibility is fundamental to science

Supporting reproducibility at the ALA

Let’s make a code snippet reproducible

Our code snippet

Our code snippet

It’s all about the setup aka a reproducible project environment

6 simple steps to make your R code run again

1. Create an R Project

1. Create an R Project

2. Use GitHub

2. Use GitHub

2. Use GitHub

Tip: Setup GitHub with {usethis}

3. Organised folder structure

4. Readable file names

4. Readable file names

5. Record packages & versions

6. Back up your data

6. Back up your data

Let’s return to our code snippet

Our code snippet

Our code snippet

Summary

Your code will run again!

On reproducibility

On reproducibility

Other useful resources

Thank you

Will your code run again?
Tips for making code reproducible in R

It’s all about the setup
aka a reproducible project environment

6 simple steps
to make your R code run again