How to make a shareable code workflow for reproducible and efficient science


Dax Kellie, Martin Westgate

@daxkellie



I acknowledge the Traditional Owners of the lands on which I live and work, the Ngunnawal people, 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.

The replication crisis

Psychology in crisis?


Of 100 replicated studies, only 39 replicated the original result

Biology & ecology in crisis?

Evolutionary biology & ecology research is not immune to wider issues of scientific research

Questionable Research Practices are prevalent but usually unintentional

Reproducibility

Available data and code?

Without knowing how someone cleaned & edited their data, even with data, reproducing a result is difficult

Available code doesn’t guarantee reproducibility

Of 62 Registered Reports from Center of Open Science database:

  • 37 had scripts available (60%)
  • 31 scripts could be run
  • 21 could reproduce main results
    (34% of total, 57% of scripts)

Reproducibility is hard

Supporting reproducibility at the ALA

galah

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

ALA Labs

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

Supporting reproducibility at the ALA

galah 


library(galah)

galah_call() |>
  galah_identify("reptilia") |>
  galah_filter(year > 2020) |>
  atlas_occurrences()



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

ALA Labs

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

The invisible workload of open research

With growing pressure from peers but little change from institutions of how research output is judged, open science is an intimidating amount of work

Small steps
to improve reproducibility*
with big impacts


*of code, mostly

1. Aim to make your work environment shareable

When loading a work environment…



This works…

rm(list = ls())
setwd("C:/Users/DaxKellie/OneDrive/Documents/ALA/Talks/ESA2023")



…but usually only on your computer

R Projects

R Projects


Without R Project

path = "C:/Users/DaxKellie/OneDrive/Documents/ALA/Talks/ESA2023/images"


With R Project

path = "./images"

The {here} package


Works nicely with R projects

library(here)
here::here()
[1] "C:/Users/DaxKellie/OneDrive/Documents/ALA/Talks/ESA2023"


Makes safe file paths easy

here::here("images", "folder", "subfolder")
[1] "C:/Users/DaxKellie/OneDrive/Documents/ALA/Talks/ESA2023/images/folder/subfolder"

The {renv} package

Initialise a new project-local R library

renv::init()

Save the state of the project to a lockfile

renv::snapshot()
{
  "R": {
    "Version": "4.2.3",
    "Repositories": [
      {
        "Name": "CRAN",
        "URL": "https://cloud.r-project.org"
      }
    ]
  },
  "Packages": {
    "markdown": {
      "Package": "markdown",
      "Version": "1.0",
      "Source": "Repository",
      "Repository": "CRAN",
      "Hash": "4584a57f565dd7987d59dda3a02cfb41"
    },
    "mime": {
      "Package": "mime",
      "Version": "0.7",
      "Source": "Repository",
      "Repository": "CRAN",
      "Hash": "908d95ccbfd1dd274073ef07a7c93934"
    }
  }
}

Use GitHub

Backup your project online, share your project with others

Use GitHub

README at the front to explain high-level context, structure, metadata

Use GitHub

Version control with GitHub Desktop is easy to learn because it’s visual

2. Readable code, readable notes

Simple, clear names


Meh

# Objects
dat_band_occsFD2 <- ...
plot2.occMod <- ...


# Functions
filter <- function(x) {...}
cool_func <- function(x) {...}


Yay

# Objects 
bandicoots_filtered <- ...
barplot_birds_high_temp <- ...


# Functions
filter_outliers <- function(x) {...}
make_map <- function(x) {...}

Simple, clear notes (with interpretations)


Meh

model <- lmer(outcome ~ predictor_1 + predictor_2 + covariate_1 + 
                (1|covariate_2) # random effect
              data = data)
summary(model)


Yay

# Test effects of temperature and rainfall on species richness
model <- lmer(outcome ~ predictor_1 + predictor_2 + covariate_1 + 
                (1|covariate_2) # random effect
              data = data)
summary(model)

# Results show significant effect of predictor_1. This suggests [interpretation]...
# However, confidence intervals of significant effect are wide

3. Render your code into a document
(with middle steps included)

Document results for easy referencing later

Quickly reference and share your work (because you don’t need to rerun your code)

Document results for easy referencing later

Quarto is like a refined, updated R Markdown - it’s easy and makes documents look nice

Document results for easy referencing later

Saving them in one place creates an accessible library of usable code.
This can be public or private

Summary

Small steps to improve reproducibility with big impacts

  • Aim to make your work environment shareable
-  Create projects with safe links (R Projects + {here})
-  Save package versions ({renv})
-  Use an online repository (GitHub)
  • Readable code, readable notes
-  Simple, clear object & function names
-  Clear notes with interpretations of results
  • Render your code into a document
-  Quickly reference your work (because you don't need to rerun your code)
-  Save rendered files somewhere findable/shareable to reference later
-  Quarto makes this easier than ever

Thank you


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

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

These slides were made using Quarto & RStudio

Slides: