Overview For Day 2

• AI & LLMs: An Illustration
• Data preparation fundamentals
• Introduction to the grammar of graphics
• Colors & accessibility

AI & LLMs: An Illustration

• We’ll use an LLM to help us prepare data for a visualization
• We’ll work with the ai_example.qmd file in the examples directory

Getting Data Ready For Visualization

• Before we can do any visualization, we need to load in data and prepare it

• Take this step seriously, especially your variable names, as your code depends on it

• Workflow options:

  1. Clean up the data in a block at the top of your notebook
  2. Separate out the data preparation from the visualization code

• There are no universal answers to the best approach here; friction as you work often tells you what to do

Option 1: Clean up the data in a block at the top of your notebook

Option 2: Separate out the data preparation from the visualization code

• In this approach, you prepare the data in a separate script or notebook

• You can then load the data into your visualization notebook

• Two options:

  1. Run the data preparation script before running the visualization script
  2. Run the data preparation script as part of the visualization script

• Example of (1) above provided in our examples/project-example directory

Benefits/Drawbacks of separating scripts

Saving Data To Load Later

• When using separate scripts, you’ll often want to save a copy of your cleaned data to load later scripts/notebooks

• A good option here is use the saveRDS function to save your data as an R data object

• This has a few benefits

  • RDS files maintain all R metadata
  • You can quickly load the data into R with readRDS, which can save you a lot of time when working with large datasets
  • RDS files are not easily to accidentally modify

Preparing Data Example

• I have created an example of data preparation in the examples/prams_1_data_prep.qmd file
• Let’s load this up in RStudio and step through it.
• Note that we’ll be opening the entire course repo as a project, so the root directory is one level up from the examples directory
• We will return to this data file later to visualize it

ggplot2 and the Grammar of Graphics

• The “gg” in ggplot2 stands for the grammar of graphics, a concept coined by Leland Wilkinson and popularized by Hadley Wickham.
• Grammar in language is a set of rules for how language is structured.
• Grammar in graphics is a set of rules for how visual representations of data are structured.

Key Concepts in ggplot2

ggplot2 workflow

ggplot2 Workflow

1. Start with data
2. Pick an aesthetic mapping
3. Choose a geometric object
4. Add statistical transformations
5. Adjust finer details: scales, coordinate systems, faceting, etc.

Aesthetic Mappings

• This is how your variables map onto the aesthetics of your figure
• It sounds highfalutin, but it usually just means:
  • What is your x?
  • What is your y?

Geometric Objects (“geoms”)

• Geoms are the elements used to represent the data
• Geoms have associated stats/functions/statistical transformations
• Stats such as counts for groups often involve aggregations

Geoms (cont.)

Position/Fill

• You can use position/fill to move elements around and color code geoms by aspects of the data, such as categories.

• For example, with a bar chart:

  • stack: Stack bars on top of each other
  • fill: Stack bars, but make them always fill up vertical space to 1
  • dodge: Put the bars next to each other

• Tip: When you reach for fill, always consider faceting in small multiples. The brain struggles with fills. More on this later.

Aggregations

• To make a bar chart, one groups data and takes a mean, proportion, or count
• Software like ggplot will try to do this for you and it works well for simple cases
• This can fail with more difficult aggregations or when doing things like applying annotation labels
• A powerful and frustration-reducing technique is to do the aggregation yourself and then use the identity mapping
• This separates the statistical logic from the visual logic and is often what makes a tricky figure easier to make

Illustration: ggplot2 Concepts

• Let’s load the prams_2_ggplot_concepts.qmd file
• In this file we will load the data we prepared in the previous file
• We will then work through the key ggplot2 concepts with examples.
• We will iterate on the figure to make it more informative and visually appealing

Illustration: Iteration & Aggregation

• The PRAMS data we used comes pre-aggregated

• In prams_3_iteration_aggregation.qmd we will illustrate how statistical transformations work by showing:

  • The identity transformation with pre-aggregated data
  • Using aggregations with simulated data

We will also demonstrate the process of iterating on a figure to make it more informative and visually appealing.

Colors

• Colors are a powerful tool for communicating information
• They can be used to highlight important information and create a visual hierarchy

RColorBrewer

Using a Color Design System

Example

• In the examples/colors_and_accessibility.qmd file, we will explore the use of color design systems and perceptually uniform colors