1. Introduction, RStudio

2. Reproducible documents with R Markdown

3. Practical

4. Advanced stuff

• Today’s slides: https://clement-lee.github.io/slides/ukrn-20251022

• Further training:
  https://bit.ly/literate-programming

• Converting from LaTeX:
  https://bit.ly/bookdown2024

• Statistical analysis: Python, MS Excel?

• Writing presentations: MS Powerpoint?

• Writing papers: MS Word, Overleaf?

• Statistical analysis: Python, MS Excel?

• Writing presentations: MS Powerpoint?

• Writing papers: MS Word, Overleaf?

• There are some caveats

• Showing you what’s possible

• Particularly useful if you have a lot of numbers / tables / figures from your analysis

• Can others run your code and get identical results?

• Can reproducibility be done in an efficient manner?

Not covered

• Can others perform the same (non-computational) experiment themselves, and reach the same conclusion with broadly the same results?

• Is my hypothesis valid? Are my methods / models useful?

• Python

https://www.theverge.com/2020/8/6/21355674

• Coding is only the first step

• Reproducibility not guaranteed in the whole process

• Wanting to revert to previous version of the analysis, but the scripts have already been updated

• Save a new set of scripts every time

• Eventually it becomes difficult to find the right version again

• Autosave & track changes help, but still not perfect

The script (.Rmd)

```{r}


plot(cars)
```
The figure above shows the distance
against speed in the cars data set.

The output (.pdf / .html)

plot(cars)

The figure above shows the distance against speed in the cars data set.

The script (.Rmd)

```{r}
#| echo: FALSE

plot(cars)
```
The figure above shows the distance
against speed in the cars data set.

The output (.pdf / .html) The figure above shows the distance against speed in the cars data set.

The script (.Rmd)

```{r}
#| echo: FALSE
#| out.width: "70%"
plot(cars)
```
The figure above shows the distance
against speed in the cars data set.

The output (.pdf / .html) The figure above shows the distance against speed in the cars data set.

The script (.Rmd)

```{r}
#| echo: FALSE
#| fig.cap: "Distance against speed in cars data set."
plot(cars)
```

The script (.Rmd)

```{r}
#| eval: FALSE

plot(cars)
```

Also useful when showing code that doesn’t work

The output (.pdf / .html)

plot(cars)

The script (.Rmd)

```{r}
#| fig.show: "hide"

plot(cars)
```

The output (.pdf / .html)

plot(cars)

The script (.Rmd)

```{r}
#| include: FALSE

plot(cars)
```

The output (.pdf / .html)

The script (.Rmd)

```{python}
import math
round(2 * math.pi, 4)
```

The output (.pdf / .html)

## 6.2832

Goals

• Think about which part of your analysis can be made reproducible

• Convert an existing script, or create a new analysis

• Generate the output in multiple format

Troubleshooting

• No laptop? Use a university computer

• Don’t know R? Some notes to get you started

• No data? Consider the following:

  • mtcars (a built-in data set)
  • https://data.ncl.ac.uk/

• Excel spreadsheet (.xlsx, .xls)?

  # do this once only
  install.packages("readxl")
  # do the following every time
  library(readxl)
  mydata <- read_xlsx("<filename>.xlsx")

A collection of .Rmd files

• index.Rmd
• 01-multivariate-data.Rmd
• 02-pca.Rmd
• 03-cluster.Rmd
• 04-regression.Rmd
• 05-regularisation.Rmd
• 06-classification.Rmd
• 07-matrix.Rmd
• 08-factorisation.Rmd
• 09-references.Rmd

index.Rmd

---
title: "MAS8383 Statistical Learning Methodology"
author: "Clement Lee"
date: Semester 1, 2023/2024  
output: bookdown::gitbook
documentclass: book
papersize: a4
geometry:
- margin=1in
fontsize: 12pt
bibliography: [references.bib]
biblio-style: apalike
link-citations: yes
---

library(bookdown)
render_book("index.Rmd", 
            output_format = c("bookdown::gitbook", "bookdown::pdf_book"))

An ongoing project

• https://github.com/ismayc/thesisdown

• Templates for different institutions

• Not yet for Newcastle University

• What if a journal has reproducibility requirements?

  • Just give them the Rmd (and data)!

• No need to give multiple scripts with convoluted instructions

• Related question: Can somebody reproduce your results on their computer?

  • Absolute paths vs relative paths
  • Package dependencies, environments

• Scope for pre-submission checks within the university?

• Package version on your machine \(\neq\) that on mine

• Changes between versions might break the code

• Goal: A clean environment for full reproducibility

# Run this once
install.packages("renv")

1. initialise {renv} in your project

   • Creates a local project library under ./renv/library

   • Creates a renv.lock file listing all package versions

   • Activates automatic isolation of your environment when you work in that folder

   library(renv)
   renv::init()

2. install packages inside the {renv} environment

   • Ensures the packages & their versions go into your project’s local library

3. write & knit R Markdown as usual

   • Local library & locked package versions automatically used as {renv} active in project

   renv::install("readxl")

4. snapshot your environment when done

   • Updates / creates renv.lock file with precise versions of all packages used

   renv::snapshot()

5. reproduce elsewhere and/or in the future

   • Installs the versions recorded in renv.lock

   • All the other machine needs is the Rmd and lockfile

   # On another machine
   library(renv)
   renv::init() # or below
   #renv::restore()

• Standalone outputs via R Markdown

  • R Markdown comes with RStudio

• Thesis / lecture notes

  • R package {bookdown} (all lowercase)
  • thesisdown for templates

• Journal articles templates

  • R package {rticles}

• Python content within

  • R package {reticulate}
  • Python itself

• Standalone output via Quarto

  • R package {quarto}

• LaTeX & packages required for pdf outputs

  • Installation outside R, or
  • R package {tinytex}

  # Run once
  install.packages("tinytex")
  library(tinytex)
  install_tinytex() # installs LaTeX (within R!)
  tlmgr_install() # installs LaTeX packages

• Full reproducibility with package versions

  • R package {renv}

• First generate the output without altering the template

• Can the output be generated and results updated if the data / analysis changes?

  • Don’t hard-code results (it beats the point!)
  • https://bookdown.org/yihui/rmarkdown-cookbook/r-code.html

• Can the output be generated on someone’s else computer?

  • Don’t include commands like install.packages()
  • Don’t use absolute paths (see right)

• Don’t include interactive commands e.g. View()

• pdf issues can usually be solved with the use of the {tinytex} package

Cons

• Initial cost (time) to convert \(\qquad\rightarrow\)

• Not as flexible as LaTeX \(\qquad\rightarrow\)

• Collaboration trickier \(\qquad\rightarrow\)

Pros

• The cost to switch increases over time

• Sufficiently developed over the years

• Possible with RStudio cloud

  • https://posit.co/blog/real-time-collaborative-editing-on-rstudio-cloud/

• Or through version control

• Coding is the first step of computational reproducibility

• Computational reproducibility is the first step of:

  • Open research
  • Useful analysis / method / model

• If you find it useful, spread the word & let us know!

• Email: clement.lee@newcastle.ac.uk

• I will send the link to the slides (or the Rmd?)

• Questions?