Image for post
Image for post
Photo by Shahadat Rahman on Unsplash

Learning data viz with D3

Diana MacDonald
May 25 · 27 min read

It was a Sunday morning in Melbourne. Having watched a few video courses on data viz with D3, I decided to set myself a goal: by the end of 2020, I will feel confident to design data visualizations of the complexity and visual impact of those by the New York Times. Today marks 30 days on that journey.

For me, data viz is a creative outlet, and a powerful intersection of design and code. It lets you tell captivating stories. The satisfaction of producing something visual, something alive with motion that you can interact with, keeps me motivated to keep playing. And because everything is new, I’m learning an incredible amount really quickly, and that fast feedback loop is compelling. Gamification is built right in.

For 30 days, I journalled in a notebook on Observable, learning out loud as I begin the journey to master this craft. You can find all my entries in this collection, Journal: Getting Started with Data Viz.

Here are the resources I used, my learning process, and a reflection on the journey so far. The second half of this story is an appendix that summarises the journal entries.

Resources

  • Learning what “good” looks like
  • Data exploration and visualization tools

Process

  • Step 1: Gather resources
  • Step 2: Data exploration
  • Step 3: Data visualization

Reflections

  • The learning curve
  • Gather resources as you go
  • Weeding out the noise
  • Avoiding mistakes
  • Finding your way in D3 docs
  • Motivating projects
  • Conclusion

About Diana MacDonald

Appendix

  • Journal entries

Resources

Learning what “good” looks like

When I teach design for coders, I encourage engineers to crank up the volume on design in their lives. Follow designers on Twitter, subscribe to newsletters, have coffee with designers at work. A lot of engineers learn to build websites from “hello world” examples with no styling, so it’s important to start improving the balance of how much quality design they’re seeing day to day. Lots of engineers also tell me they don’t “have an eye for design” or don’t know what sources to trust because “design is subjective”, so it’s helpful to have a hand in changing that.

Here are some of the industry legends in data viz recommended to me. Their works have helped me learn what “good” looks like in data viz:

And some folk to follow on Twitter:

Photo of Nadieh Bremer’s Hubble’s 30-year legacy project showing scientific observation data stylised as a constellation
Photo of Nadieh Bremer’s Hubble’s 30-year legacy project showing scientific observation data stylised as a constellation
Hubble’s 30-year legacy by Nadieh Bremer

Data exploration and visualization tools

During the first 30 days, I explored data in Vega-Lite and then levelled up with a D3 visualization:

  • Vega-Lite: a visual grammar for specifying charts in JSON. It’s convenient for quickly switching between “marks” (bar, area, point, etc.) with a single underlying data set to get a sense of the shape of the data.
  • D3: A JavaScript library for visualizing data with HTML, SVG, and CSS. It’s powerful and works seamlessly on the web. It’s a swiss-army knife for visualising data in digital products, making it a great tool for web developers.

Moving forward, I want to skip Vega-Lite and instead use Data wrapper. It’s a much faster method to see your data, while D3 gives you far greater control over the final results.

To search D3 documentation by method name and so on, I use Dash.

Alfred search “- d3: domain” results: time, point, ordinal, quantile, quantize, diverging, threshold, continuous & sequential
Alfred search “- d3: domain” results: time, point, ordinal, quantile, quantize, diverging, threshold, continuous & sequential
I use Dash’s integration with Alfred to quickly search D3 docs

Process

Data visualization has a wildly steep learning curve. It’s multidisciplinary, requiring expertise in data, design, and development. Within each, you need to learn the theory and tools, as well as practice the craft.

To start with, I wanted to focus on only the development step. This is a challenging place to start — you cannot build a visualization without data or design. This may have made it more difficult for myself, but it helped me get to a tangible outcome faster: produce charts.

While building my first bar chart, I needed to learn the minimum amount about Observable itself and data wrangling.

Observable:

  • Observable user manual. If you’re learning Observable, I suggest you read at least the first two sections. Side note: Observable is hands down the best live coding environment I’ve used. Blows codepen, JSfiddle, and all the others out of the water. It’s snappy and reliable. You can edit a cell, run it, and see the result instantly.

Data wrangling:

Screenshot of Observable’s User Manual with an illustration of a child reading a book
Screenshot of Observable’s User Manual with an illustration of a child reading a book
Observable’s user manual is itself an Observable notebook linking to info about its awesome features

Step 1: Gather resources

For each chart type I wanted to learn, I’d start with learning some theory. I’d focus on the function of the chart, such as comparison, concept visualisation, correlation, distribution, geographical data, part to whole, or trend over time. I’d also consider the appropriate data types for them and look out for caveats. Finally, I’d compare them to similar charts, such as differentiating line charts from sparklines or scatterplots in 10 May 2020: Line charts.

For learning the theory:

Standing on the shoulders of giants, I would then check out visual examples, code examples, and examples in design systems.

For visual examples:

For code examples:

For design systems:

For any remaining questions I had, I would search Slack archives:

Even if you’re not into live chat, these Slack communities are a treasure trove of questions answered by industry experts, so you can treat them like a modern and kind Stack Overflow for data viz.

Screenshot of Gallery notebook showing 20 thumbnails of visualizations in other notebooks
Screenshot of Gallery notebook showing 20 thumbnails of visualizations in other notebooks
The Observable D3 Gallery by Mike Bostock is an amazing resource

Step 2: Data exploration

For each chart type, I would assemble a dataset appropriate to that chart type. Either I would collect my own or search for a beginner-friendly dataset (that is clean and not missing data):

To see the shape of the data and learn a data exploration tool, I’d build each chart type in Vega-Lite:

Screenshot of Vega-Lite homepage
Screenshot of Vega-Lite homepage
Vega-Lite is a JSON specification format for rapidly creating charts

Step 3: Data visualization

After gathering resources and exploring data, I built charts in D3. Generally, I’d find a handful of working notebooks on Observable and recreate them by hand from memory instead of forking, to ensure I understood every line.

For learning D3, I recommend reading these tutorials:

For more resources, see my Links notebook on Observable.

Image for post
Image for post
How to learn D3.js by Amelia Wattenberger uses plain English and clear visuals to teach D3

Reflections

The learning curve

Beyond data viz being a multidisciplinary field, learning to wield D3 requires advanced knowledge of SVG, JavaScript, and CSS, and a healthy amount of patience. Otherwise, it can be challenging to learn these web languages at the same time while attempting to debug and overcome setbacks.

I’ve tackled projects with steep learning curves before, though one notable difference is the amount you need to learn in data viz before you can reliably produce a single, basic visualization. With most other projects, I’ve worked quickly to skill up enough to complete the first, end-to-end realistic piece of work, before continuing on to iterate, refine, practice and expand on that end-to-end piece. For example:

  • In music, you might focus on slowly playing a song correctly, before building up speed.
  • In product design, you might focus on producing a minimum viable product, then evolve from there.
  • In web development, you might focus on scaffolding an app and deploying it to the Internet, then building it out from there.

In data visualization, there’s very little you can achieve without first grasping data collection, cleaning, and other wrangling. You must also know how to code, and learn a new library. With D3, especially, you need to know exactly what you’re trying to achieve with every mark, axis, label, legend, and interaction. This means applying visualization design principles at the same time as coding.

Little strokes fell great oaks

It’s easier to tackle a huge task with steady and persistent effort. Aim for 1% better everyday. I made this work for me by aiming to publish a notebook everyday. There was no minimum length or time required (though I am sure it would feel silly to publish a single sentence). Sitting down to write 1 notebook and otherwise change 1 line of code was enough to keep data viz top of mind for 30 days. But it’s easy to focus on the first five minutes to get yourself started.

Screenshot of Observable journal collection showing the first 9 notebooks with slightly varying bar charts
Screenshot of Observable journal collection showing the first 9 notebooks with slightly varying bar charts
My daily journal led to small, consistent effort… and then not wanting to break the chain

Gather resources as you go

Along the way, I published useful Links in Observable.

Whenever I stumbled across a link that might be useful in the future but I wasn’t ready to look at it yet, I collected it in my private notes. When the day arrives that I want to build maps or learn Python, I already have great resources to fast-track that effort. Once I’ve road tested them, I’ll add them to my published links notebook.

Screenshot of ObservableReference collection includes links, working with Observable, working with Vega-Lite, and a sparkline
Screenshot of ObservableReference collection includes links, working with Observable, working with Vega-Lite, and a sparkline
I regularly refer back to my reference collection

Weeding out the noise

D3 changed a lot from v3 to v4, and again some to v5. At this time, the Internet is still chockers with tutorials and blog posts for older versions. Some key things to look out for:

Scatterplot of exercise in April–May 2020
Scatterplot of exercise in April–May 2020
This scatterplot shows a voronoi overlay for triggering tooltips

Avoiding mistakes

Where possible, learn from Mike Bostock’s notebooks. They’re the most up-to-date notebooks you’ll find and follow lots of great conventions that will save you headaches, even before you understand the reasons for them. One drawback is that these SVGs don’t appear to have been designed accessibly.

What to use in your own notebooks:

  • Use a specific version of D3 (or other modules) in imports e.g. d3 = require("d3@5")
  • Use generic, conventional variable names, such as x, y, xAxis, yAxis, margin, height, and data. This makes it easy to adapt other people’s code to your needs.
  • Similarly, rename columns or row headers from your data when you load it in so you can reuse your own chart code with different datasets and only update your code in 1 place (the data cell) instead of every cell where the data field names are used. For example: Object.assign(d3.csvParse(await FileAttachment(…).text(), ({MyDataTitleField: name, MyDataXField: x, MyDataYField: y}) => ({name, x: +x, y: +y}))). Learn more in 23 May 2020: Area Charts in Vega-Lite and CSV Parsing.
  • Pin a data[0] cell to remind yourself what the data looks like.

What to look out for when learning from other people’s notebooks:

  • Check the D3 import in the appendix at the end to see if it’s pulling in other modules.
  • Check the data cell to see what data wrangling work might be necessary and the shape of the data.
  • Check the appendix for anything else unusual.

Some debugging tips I’ve figured out:

  • To console.log() something in the middle of a chain of method calls, you can return console.log(variable) || (notice the logical OR pipes ||). For example: .join("circle").attr("stroke", d => console.log(d) || d.r)
  • To test the behaviour of your scale functions, first try to run it with any data like x(123) instead of intended usage, x(d.x)
  • Check your brackets, e.g. .range([height - margin.bottom, margin.top]); (one challenge of Observable is debugging when silly stuff happens like forgetting brackets)
  • Give elements that you append IDs or classes to make them easier to identify during debugging and to make sure they don’t clash with other things on the page (if you put your chart on a page with other elements). That is, don’t depend on broad selectors, such as d3.selectAll('path').
Area chart showing 863 cumulative cells written in 28 Observable notebooks in April and May
Area chart showing 863 cumulative cells written in 28 Observable notebooks in April and May
By May 23rd I’d written roughly 863 paragraphs or code blocks

Finding your way in D3 docs

D3’s documentation is written in a style that’s not intuitive to me. Here are some hints to give you the lay of the land:

  • Many functions are overloaded. This means the function will do one thing with no arguments, something else with 1 argument, something else with an argument of a different type, and so on. You may need to scroll up and down the docs to find adjacent explanations for a function called with different arguments.
  • To learn about, for example, .domain() on a scaleLinear(), you need know that scaleLinear()is a type of continuous scale (along with power, log, identity, time, and radial), so you need to search for continuous.domain()
  • To learn about, for example, .on(), you need know that it’s used on a selection, such as d3.select(), d3.selectAll(), d3.create(), so you need to search for selection.on(). In examples, it will also likely appear after a selection’s “transformation method”, such as selection.attr(), selection.classed(), selection.style(), selection.text(), selection.html(), selection.append(), or selection.insert().
  • There are loads of aliases and the like, so reading the docs can mean bouncing around a lot, following links to “equivalent” functions.
  • There’s a bit of jargon to wrap your head around and it’s often not linked to a definition. Sometimes the jargon isn’t explicitly defined anywhere, just sort of inferred. I found myself occasionally rewriting documentation to strip out the jargon so I could make sense of it. As an example, compare the line.defined() documentation with my explanation of line.defined() in 12 May 2020: D3 Line Chart.
  • Sometimes the syntax is buried within prose, but it’s worth digging for. I’d prefer to see the parameters and return values called out more prominently, like I’ve seen in documentation on MDN.

I’m under the impression there’s a huge amount of work under way to migrate old tutorials and examples from bl.ocks.org to Observable. Some folk are putting in an incredible effort to ease the learning curve into D3, and I’m excited about where it’s headed.

Hovering over each line shows that country’s life expectancy, including Rwanda at 24 years during the Rwandan Civil War
Hovering over each line shows that country’s life expectancy, including Rwanda at 24 years during the Rwandan Civil War
This multi-line chart with annotations shows the effect of the Rwandan Civil War on life expectancy

Motivating projects

Given the steep learning curve, I’ve heard a few folk talk about finding motivating and manageable projects. In this round up, Mike Bostock to humans: ‘Try to look for small problems first’ by Sérgio Spagnuolo on the Mike Bostock, Reddit AMA, there were a few quotes that stood out to me:

Try to look for small problems first, the sort of thing you can solve once per day, whenever you have time. The rewards from early victories are strong motivation to keep going.

… think of small coding problems that you are comfortable solving, and then increasingly ramp up to larger problems as you go. The satisfaction you derive from solving the smaller problems will motivate you to keep going.

I find it to be the easiest thing in the world to work on something if you are passionate about it, and you can break it up into small pieces (like examples) that you can publish and share with others for external validation. So probably, choosing to work on things you are excited about, and then finding space to avoid distractions or interruptions is the key.

Along a similar vein, Josh Temple wrote in Why I abandoned online data courses for project-based learning:

The key is to find a project that you’re so excited to bring to reality that you don’t mind pushing through the obstacles along the way. Here are four attributes that I find make a project motivating. Maximize one or more of these and you’ll have a much better chance of actually finishing your project.
- Utility: “This would make my life easier and save me time.”
- Passion: “This would solve a problem I deeply care about.”
- Curiosity: “This dataset really intrigues me and I want to explore it more.”
- Competition: “I want to win this prize and beat other competitors.”

For my part, I’m yet to tackle a large and meaningful project, so the greater challenges are persevering through bugs and maintaining momentum. I’ve mostly focused on tackling hard problems over the weekend with the time, space, and focus to work through thorny problems or track down documentation and solutions. On work days, I’ve scaled back my efforts to focus on learning theory or tweaking the previous day’s chart in some small way.

Once I have a good handle on the basic tools, I will be deciding on a lovable project. I am noodling with some ideas already.

A donut chart of small batch scones ingredients, Buttermilk, Cold butter, and 62.5g Self-raising flour, from Sarah Cooks
A donut chart of small batch scones ingredients, Buttermilk, Cold butter, and 62.5g Self-raising flour, from Sarah Cooks
Small batch scones are my favourite part of lockdown so far

Conclusion

And that’s a wrap. In the appendix below, you can find more details about what I covered and lessons learned each day in my journal.

Otherwise, I only want to add that I’m grateful for the selfless people who have spent countless hours creating tutorials, books, and courses, and making them available for free. I’m inspired by the passionate creative coders who make beautiful works that have impact in the world, and share them everyday. Thanks so much for all that you do.


About Diana MacDonald

Diana MacDonald is the author of Practical UI Patterns for Design Systems and creator of Typey Type for Stenographers. She leads the design systems team at Culture Amp. Raised in the tropical north of Australia, she has spent the last decide in the tech industry, exploring the digital space with progressive organisations like Culture Amp, Bellroy, and SitePoint. Blurring the lines of designer and developer, she believes in the value of considered, inclusive, and remarkable stories. She wants to help you effortlessly execute your digital ideas.


Appendix

Topics covered and lessons learned each day in Journal: Getting Started with Data Viz.

Day 1–7: Learning the basics with bar charts

25 Apr 2020: First Bar Chart

26 Apr 2020: Bar Chart Revisited

  • Setting .padding(0.1) reduces the width of each bar rather than adding space between each bar. This makes sense for conveniently working with bars and increasing padding without blowing the whole chart out.
  • .tickSizeOuter(0) removes the stump on the bottom left of the chart (it hides the first tick on the axis)
  • .nice() ensures the top of the chart finishes at 13% instead of 12% with the highest bar poking above the highest tick.
  • width is the current width of a cell and is part of Observable’s standard library. It is a Reactive variable that instantly responds to changes in the window size.
  • “Use caution when applying global styles: if a style affects a cell’s height, the runtime may not notice unless the affected cell is re-evaluated.” — Introduction to HTML
  • Some numbers can’t be represented in binary floating-point, meaning code like tickFormat(d => (d * 100) + "%") can produce “7.000000000000001%” instead of 7%. You could force a “fix” using .toFixed([digits]), or round it using Math.round(yourNumberHere * 10 ) / 10, but use the better option, d3.format(".1f") from the d3-format module.
  • Showing the median as a line when it’s one value or highlighting bars when it’s two.

27 Apr 2020: Finalising the Bar Chart

  • The general term for lines showing the median, mean, benchmarks, or “best fit” is “reference line” (or “reference band” for showing a range).
  • Use .attr('dy', '1em') (specifically 1em) to shift text along its y-axis by the height of the text.
  • Set styles using .style('font', '400 12px/1.5 "Work Sans", "Open Sans"') instead of .attr('style', 'font: 400 12px/1.5 "Work Sans", "Open Sans"')
  • Use Object.assign(target, ...sources) to add labels for axes as properties to the data object after parsing it. Add the await keyword to ensure the data promise is resolved before adding the new properties to it.
  • There is no translateY SVG transform (it’s a CSS transform). Use translate instead e.g. .attr('transform',translate(456, 0))
  • Styled axes by Mike Bostock. Use a right axis with tick sizes that match the length of the chart to push it to the left side of the chart.
  • Mike Bostock’s Stack Overflow answer to styling axes
  • D3 chart title.
  • D3 axes labels.

28 Apr 2020: What Else I Learned From That Bar Chart

  • Specify a specific version of D3 in the appendix to future-proof against breaking changes: d3 = require("d3@5")
  • band.range([range]) … sets the scale’s range to the specified two-element array of numbers.
  • band.rangeRound([range]) Sets the scale’s range to the specified two-element array of numbers while also enabling rounding. This will “give the results as integer values, using Math.floor() to avoid overflowing the range. This will usually leave some unused space, that must be allocated to the left and right padding (even if those have been set to 0). Align them with band.align.” —d3.scaleBand on Observable

29 Apr 2020: Setting up Observable

  • Interactions: Drag the slider to set the exit transition duration on mouseover/hover.
  • Use .align([align]) on bands to left all the bands to the left (0), center (0.5), or right (1).
  • Observable view.

30 Apr 2020: What Else Can I Do With This Bar Chart?

  • Reorder adding SVG elements so they appear in the correct order on screen. For example, ensure the bar value labels appear above the median line for readability.
  • A CSS hack to give labels some distinction: apply a white text shadow in 4 directions using .style("text-shadow", "white 1px 1px, white -1px -1px, white 1px -1px, white -1px 1px")
  • D3 bar labels.

1 May 2020: Formatting Numbers and Sorting a Bar Chart

  • D3 Format to format numbers the way you like using d3.format. This is a longer explanation of the format specifier.
  • Use JavaScript’s array.sort() to order bars before binding them.
  • D3 grouped bar chart with color.

Day 8–11: Data wrangling and exploration with Scatterplots

2 May 2020: Exploring More of Observable and Vega-Lite

  • Use HTML tagged literals to add CSS style tags: “html<style> .highlight { background: yellow; } </style>
  • Read Observable’s Introduction to HTML by Mike Bostock to learn about reactive Markdown, “${tex\KaTeX}” for math, sparklines for a chart inline with prose, using canvas, using D3, using SVG tagged template literals, using viewof to react to an input element’s changing value, and adding global CSS using html tagged template literals.
  • Read Observable’s Introduction to Promises by Mike Bostock to learn about Observable’s delay, tick and when functions.
  • Read Observable’s not JavaScript by Mike Bostock to learn some “gotchas” to look out for working with Observable, such as “you’ll need to wrap object literals in parentheses or use a block statement with a return” e.g. object = ({foo: "bar"}). It also introduces Observable’s “special mutable operator so you can opt-in to mutable state: you can set the value of a mutable from another cell”. Finally, it talks about how to import things across notebooks.
  • “inlineCode” helper to add some custom inline code styling in Observable.
  • D3 sortable bar chart.
  • Vega-Lite scatterplot.

3 May 2020: Exploring More Vega-Lite

  • I published a fork: Vega-Lite API Vega-Lite V4. This way I can intentionally import Vega-Lite major version 4 into my notebooks. When Vega-Lite major version 5 comes out with non-backwards-compatible breaking changes, my older notebooks using this import won’t randomly break. This future-proofs my notebooks using Vega-Lite V4.

4 May 2020: Vega-Lite Scatterplot and Heat Map, and More Posts on Getting Started With Data Viz

  • Read Observable’s Introduction to Imports by Mike Bostock to learn a bit more about Observable imports, like how imported cells are lazily evaluated and you can import from private notebooks, even though that could cause broken behaviour in a published notebook.
  • Use ~~~ for code fences in Observable including tagged code fences: ~~~js
  • The SVG hyperlink element <a> is a container around any shape. If it’s around a circle, it will have a circle-shaped hover/click target.
  • In Observable, a view is “a user interface element that directly controls a value in the notebook”. It has two parts: a view (typically an interactive DOM element) and value (any JavaScript value). See also: Introduction to Views by Mike Bostock.
  • It seems that a viewof is a second hidden cell that shows the current value of a user interface input element. See: A Brief Introduction to Viewof by Mike Bostock.

5 May 2020: Vega-Lite Scatterplot continued

  • Scatterplot theory.
  • Vega-Lite proportional area (bubble) charts.
  • Vega-Lite color encoding.
  • Vega-Lite rows.
  • Vega-Lite quantize scale.
  • Heatmap theory.
  • Vega-Lite Heatmap.
  • Tidy data is where:
    1. Each variable forms a column
    2. Each observation forms a row.
    3. Each type of observational unit forms a table.

Day 12–15: D3 Scatterplots with legends and tooltips

6 May 2020: D3 Scatterplot

  • Vega-Lite tooltips.
  • Vega-Lite grid removed.
  • Vega-Lite line mark interpolation.

7 May 2020: D3 Scatterplot with Legends

  • d3-time-format
  • Use man strftime on the command line to figure out the arguments for the d3.timeParse()function so that “%d” is replaced by the day of the month as a decimal number (01-31) and “%b” is replaced by national representation of the abbreviated month name.
  • Renaming fields upon parsing CSV data file. There’s a longer post about CSV parsing in 23 May 2020: Area Charts in Vega-Lite and CSV Parsing.
  • A JavaScript Map for color const exerciseColorMap = new Map([["Bike", "#8242a8"], ["Run", "#ff1493"], ["Walk", "#FFCE1E"]]) to conditionally set the data points stroke color according to its type.
  • Read scale.ticks by Fil to learn more about ticks. Roughly, .ticks() gives us the array of some values from the scale’s domain while .tickFormat() takes the same arguments and gives us the formatting for those ticks. This kind of detail is really helpful for making sense of the API: “The tick format provided by time scales ignores the specified count; the returned string is based solely on the given time. This is sometimes surprising, but allows the format to behave consistently across views when the domain changes, improving readability during animated transitions and zooming.”
  • A screenshot of creative commons licensing used in a notebook.
  • D3 scatterplot with legend

8 May 2020: D3 Scatterplot with Tooltips

  • Read about Observable’s Introduction to require. You can have multiple inputs at once such as: d3CsvAndFetch = require("d3-dsv@1", "d3-selection@1")
  • Using Mike Bostock’s color legend, add swatches as a separate cell: swatches({ color: d3.scaleOrdinal(["Bike", "Run", "Walk"], ["#8242a8", "#ff1493", "#FFCE1E"]) })
  • Choropleth with Tooltip by Duy Nguyen shows an example of a legend presented on a chart using the same color legend notebook
  • Import Susie Lu’s d3-legend using: d3Legend = require('d3-svg-legend'). Notice the ‘-svg-’ in the module name there unlike the name of the project.
  • D3 scatterplot with legend
  • Selection join by Mike Bostock shows a join example with separate enter and update behaviour.
  • D3 scatterplot with tooltips using Line Chart with Tooltip by Mike Bostock, callout and bisect.
  • I added a crude fix, if (!b) { return a; }, to prevents errors when mousing over right edge of chart. These errors appear in the original notebook.
  • SVG has its own span element for styling parts of text elements: `tspan`.
  • Read MDN’s toLocaleString documentation
  • Voronoi Tooltip by Ajayyy shows notebook appendix formatting that chunks out “setup”, “scales”, and “data”. It also adds cells specifically for “plotAreaWidth” and “plotAreaHeight” to substract margins, which keeps some of that noise out of the chart cell: plotAreaWidth = width - margin.left - margin.right
  • Expand d3 imports with other modules: d3 = require("d3@5", "d3-delaunay@5")

9 May 2020: D3 Scatterplot with Voronoi Tooltips

  • “Use transparency to help with overlapping data points”.
  • Read this in-depth tutorial on Using a d3 voronoi grid to improve a chart’s interactive experience by Nadieh Bremer to understand the intent and usage of voronoi in charts, but then read Learn D3: Interaction by Mike Bostock to learn how to use d3-delaunay to implement polygons for tooltip trigger regions.
  • It looks like d3-delaunay replaces d3-voronoi completely, even though d3.voronoi appears to be built into d3 and recommended in the API docs while d3-delaunay appears to be separate. The d3-voronoi repo shows this message: “Deprecation notice: Consider using the newer d3-delaunay instead of d3-voronoi. Based on Delaunator, d3-delaunay is 5-10× faster than d3-voronoi to construct the Delaunay triangulation or the Voronoi diagram, is more robust numerically, has Canvas rendering built-in, allows traversal of the Delaunay graph, and a variety of other improvements.”
  • Links to other articles and notebooks about using voronoi.
  • Voronoi tooltips with overlay.
  • Note: you don’t need the SVG overlay. It may be possible to use delaunay.find() directly without the overlay as shown in Summer heat 🍦 🌞 by Fil.

Day 16–17: Line chart theory and exploration

10 May 2020: Line charts

  • Line chart theory
  • Spline graphs, monotone functions, curving functions, and oversampled data.
  • Don’t “cut” the Y Axis
  • Spaghetti plot
  • Dual axis
  • 11 May 2020: Vega-Lite Line Charts
  • Vega-Lite line charts
  • “You have to be like the worst tabloid newspaper in the front and the Academy of Science in the back” — Hans Rosling
  • Covid Tracker Tracker [sic]

Day 18–21: D3 line charts with annotations and shading

12 May 2020: D3 Line Chart

  • D3 line chart with 1 series
  • A Y Axis label can double as a chart subtitle. Here, I’ve used “Australian life expectancy in years” as the Y Axis label and chart title.
  • Add classes to SVG groups just to name them, so you can easily find and navigate them when inspecting them in the console.
  • Different “nice” formatting: .domain(d3.extent(data, d => d.year)).nice(d3.timeYear.every(2))
  • Use the line function’s “defined” accessor: .defined() to check if there are any data points missing. If so, skips the missing line segments. This is a longer explanation of “defined”. This is the line generator code: line = d3.line().defined(d => !isNaN(d.life_expect)).x(d => x(d.year)).y(d => y(d.life_expect));
  • Examples of cutting the Y Axis
  • Sparklines, including inline canvas sparkline
  • “The only function of economic forecasting is to make astrology look respectable.” — John Kenneth Galbraith

13 May 2020: D3 Multi-Line Chart

  • Reformatted data cell to include a time series array of names and associated values (series) and the list of years as an array (dates).
  • The console showed an error: Uncaught TypeError: d3.least is not a function. This uses d3-array: d3 = require("d3@5", "d3-array@2").
  • The subsequent error, y = TypeError: t is not iterable, revealed that my time series values data was a single value instead of an array of all the values for that series.
  • Multi-line chart with dot labels and hover to emphasise 1 line at a time using bisectLeft and moved, entered, and left methods.

14 May 2020: D3 Multi-Line Chart with Annotations

  • D3-annotations: d3 = require("d3@5", "d3-array@2", "d3-svg-annotation@2")
  • Example of d3.annotationCallout
  • The annotation note title also lightens when hovering over other lines.

15 May 2020: D3 Multi-Line Chart with Shading

Day 22–24: Pie and donut chart theory and exploration

16 May 2020: Donut Charts and Pie Charts

  • Pie chart and donut chart theory

17 May 2020: Pie Charts in Vega-Lite

  • Vega-Lite pie chart of scone ingredients
  • Errors due to versions: It looks like the example in Hello, Vega-Lite by Mike Bostock is using an older version of Vega-Lite. The notebook from which we’re requiring vegalite appears to be an old version of the notebook.
  • Use vegaEmbed = require("vega-embed@5")
  • Use an embed function embed = function(spec) { return vegaEmbed(spec, {loader: vegaEmbed.vega.loader({baseURL: 'https://vega.github.io/vega/'}), actions: true}) }
  • Wrap data keys in double quotes, because JSON demands strings for keys.
  • "view": {"stroke": null}, removes the chart border
  • "legend": null, removes the category color legend in "color": {"field": "category", "type": "nominal", "legend": null}
  • It looks like "stack": true places the labels correctly. The Vega-Lite stack docs focus on stacking bar charts and only mention once that it could be applied to the theta channel. It does not describe what it means to have things “stacked” with a theta channel so I will just file away in my brain as “the thing that does what I need with pie chart labels”. After a bit more reading I spotted “For now, you need to add stack: true to theta to force the text to apply the same polar stacking layout.” in Pie Chart with Labels
  • Use , "stroke": "#fff" to the arc mark to distinguish sectors.
  • Use "radius": {"field": "quantity", "type": "quantitative", "scale": {"type": "sqrt", "zero": true, "range": [40, 100]}}, to create a Vega-Lite radial plot.
  • As far as I can tell, the default sort order is ascending but that will only work on a flat array of data using the specific key values. My data uses an array of objects, so I needed to add this line to my top-level encoding: "order": {"field": "quantity", "sort": "descending"}
  • Vega-Lite tooltips: "tooltip": [{"field": "category", "type": "nominal"}, {"field": "quantity", "type": "quantitative"}]
  • Chart title ("title": {"text": "A Scone", "dy": -12},) and padding ("padding": {"top": 24, "right": 24, "bottom": 24, "left": 24},)
  • Chart colors: "color": { "field": "quantity", "type": "nominal", "scale": { "range": ["#FDBAAB", "#FFEBA5", "#99C3E1", "#90D1C5"] }, "legend": null },

18 May 2020: Donut Chart in Vega-Lite

  • To change a Vega-Lite pie chart to a donut chart: "mark": {"type": "arc", "innerRadius": 100, "outerRadius": 150, "stroke": "#fff"}
  • Add arbitrary text using a text mark and a value key instead of a field. See the Vega-Lite textdocs and the Vega-Lite value docs.
  • Give the arc its own encoding and tooltip to avoid tooltips over the center of the donut.
  • In Observable, give images alt text: { const image = await FileAttachment("image.png").image(); image.alt = "Screenshot of donut chart with open cell clipping tooltip." image.width = 375; return image; }

Day 24–27: D3 donut charts with footnotes and styled tooltips

19 May 2020: Donut Chart in D3

  • Choose Sequential, diverging, categorical or cyclical color scales as appropriate for the data.
  • “There is no difference between [line or thick arcs], other than the thinnest donut being worse than the rest (we’re not sure exactly why)” in The humble pie chart: part2 by John Nixon, Office for National Statistics.
  • Experiments with different color schemes and approaches to adding them to range: .range(d3.schemePastel1), .range(["#8242a8","#cf3f93","#fc5972","#ff8a51","#ffbf40"]), .range(d3.quantize(t => d3.interpolateCool(t * 0.8 + 0.1), data.length)), .range(d3.quantize(t => d3.interpolateSpectral(t * 0.8 + 0.1), data.length).reverse())
  • Color resources.
  • D3 donut chart.

20 May 2020: Donut Chart in D3 with Footnotes

  • To add a “source” line of text, append a text element and set the innerHTML using .html() from D3’s selection.html.
  • Alternatively, dump all the source details inline in a .html() call.
  • Use a figcaption element to add notes and source: <figcaption style="padding-bottom: 12px; text-align: center; font-family: 'Work sans'">Source: <a href="https://www.sarahcooks.com.au/2020/04/small-batch-scones.html" style="fill: #3c72d7;">Sarah Cooks: Small batch scones</href></figcaption>
  • Wrap the chart and figcaption element in a figure element: <figure style="max-width: 100%;">${chart} ${caption}</figure>
  • “Avoid chart footnotes where possible. If extra information is needed: annotate the chart, include the information in the statistical commentary accompanying the chart, add a footnote to the chart title”
  • Some notebooks showing source or notes.
  • The SVG title element can also be used on an “graphics element”, including <circle>, <ellipse>, <image>, <line>, <mesh>, <path>, <polygon>, <polyline>, <rect>, <text>, and <use>. As with title attributes used on HTML elements, this produces shabby tooltips if you hover for long enough.

21 May 2020: Donut Chart in D3 with Tooltips

  • D3 donut chart with tooltips.
  • The approach here uses HTML for the tooltip itself and translates it into position using CSS according to the data passed to the onmouseover function from the D3 selection.on() listener on the arcs. The advantage of using HTML and CSS is that the tooltip automatically resizes based on the width of the text content inside it.
  • I have two working theories on the absence of standard conventions or libraries for styled tooltips in D3 charts:
    1. Every visualisation is unique and D3 and SVG both give you such control that you might as well tailor your tooltip to every situation. Compare the 3 examples in AntV’s G2 Visualization Grammar Tooltip examples.
    2. Data viz folk generally use other methods to let people access “details on demand”, such as showing values directly on data on hover (especially for bar charts).
  • d3-selection: “By convention, selection methods that return the current selection use four spaces of indent, while methods that return a new selection use only two. This helps reveal changes of context by making them stick out of the chain”

Day 28–30: Area charts and CSV parsing

22 May 2020: Area Charts, Stacked Area Charts, Stream graphs and Ridgelines

  • Theory about area charts, stacked area charts, stream graphs and ridgelines

23 May 2020: Area Charts in Vega-Lite and CSV Parsing

  • Vega-Lite area chart showing 863 cells written across 28 Observable notebooks.
  • Using CSS counter to count cells in notebooks via the minimap: element.style { counter-reset: count 0; } .minimap-cell-row:before { content: counter(count)"."; counter-increment: count; }
  • Wrangling CSV data
  • Observable file attachments: Calling attachment.text() returns a promise to a string. We then wait for this promise to resolve before calling d3.csvParse.
  • await waits for a Promise.
  • From the d3-dsv module, d3.csvParse calls dsvFormat(",").parse() with a comma as the delimiter, which constructs a new DSV parser and formatter. Just below dsvFormat in the docs, you can see dsv.parse(string[, row]), which tells us about the DSV parser.
  • Row conversion function, parameter and body.
  • Instead of d in our cell, you can see an Object using ES6 destructuring assignment syntax to assign the data row’s name (e.g. “Number of cells”) to a local JavaScript variable (e.g. y).
  • Our d destructuring assignment is wrapped in parentheses because they’re “required when using object literal destructuring assignment without a declaration.”
  • The => indicates an ES6 arrow function expression using the syntax, (param1, param2, …, paramN) => expression.
  • The docs show that destructuring within the parameter list is also supported as advanced syntax, which is what we’re doing.
  • “In a concise body, only an expression is specified, which becomes the implicit return value. In a block body, you must use an explicit return statement.” In order to use a “concise body” to return our object literal, we wrap our expression in parentheses.
  • From the d3-time-format module, we then use d3.timeParse, which is an alias for locale.parse(on the default locale), to create a time parser: const timeParser = d3.timeParse("%d %b %Y");
  • This time parser expects a date that looks like “23 May 2020”. We use it to parse the “Date” as our X value: x: timeParser(x),
  • The + in the line y: +y is an Arithmetic operator: Unary plus (+) that “attempts to convert it into a number, if it isn’t already”.
  • We add extra JavaScript properties to our data array to store our X Axis and Y Axis labels: const extraPropertiesSource = {xAxisLabel: "Date →", yAxisLabel: "↑ Cells"}; return Object.assign(dataObjectTarget, extraPropertiesSource);
  • This uses Object.assign(target, …sources) to copy all the (enumerable own) properties from one or more source objects to a target object and return it.
  • Experimenting with different timeUnit values: yearmonth and utcmonthdate
  • The Vega-Lite time unit docs explain that “The specifier monthdate is sensitive to month and date, but not year, which can be useful for binning time values to look at seasonal patterns only.” and to “use UTC time, you can add the utc prefix (e.g., "utcyear", "utcyearmonth").
  • Read Vega-Lite axis to learn about "titleAngle": 0 to rotate the Y Axis label and "titlePadding": 25 to pad it from the tick markers.
  • Using labelExpr described under Labels in Vega-Lite: Axis documentation and the example shown under Example: Using Axis labelExpr to Display Initial Letters of Month Name, I can show the first letter of the X Axis ordinal datum label using "labelExpr": "datum.label[0]". This is a bit useless here, but it may come in handy in the future.
  • Using Vega-Lite condition: "condition": {"test": "datum['cumulative_count'] > 850", "field": "cumulative_count"},

24 May 2020: Area Charts in D3 with Tooltips

  • Bisect for tooltips.
  • Redrawn SVG path for a callout with the tip underneath the tooltip.
  • SVG path visualizer by Mathieu Dutour
  • SVG path editing to include variables to adapt to the width of the text label’s bounding box.
  • In Radial Area Chart by Mike Bostock, the text labels appear to have a white shape around them, achieved using duplicated text elements using a large white stroke behind the actual text.
  • D3 area chart with tooltips.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch

Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore

Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store