Visualizing my Reading with Semiotic
When I tracked my reading during 2019 I had a loose idea that I might want to do some simple data visualization to look back on the year. Ask any designer or developer and this can quickly devolve into endless yak shaving, and indeed there was a moment where my brain was like, “Clearly I need to learn React Fiber and D3”. Thankfully I stepped away from that larger goal (distraction?) and found Semiotic, created by Elijah Meeks. Meeks has written a lot about dataviz, React, and D3 and together with Amy Cesal and Mollie Pettit cofounded the Data Visualization Society.
Semiotic provides a lot of common chart types out of the box, including what I was looking for: a timeline. I used that as a jumping off for my visualization. Here’s the in-progress result, if you want to skip ahead.
Dates and Time Scales #
The Semiotic Timeline example is handy, but my timeline only stretched across one year. So I had to figure out how to map a different time scale to the chart.
Adding Month and Day information #
I added month/day information as a JavaScript date. Here’s what a book looked like in my first pass at the data set
{
title: "Conversations with Friends",
author: "Sally Rooney",
start: new Date(2019, 6, 27),
end: new Date(2019, 6, 29),
genre: "Fiction"
},
A few things to note:
- This worked, but if I wanted to externalize my data into a JSON file this would not be valid
- Dealing with dates this way you run into the weirdness of having the month be 0-indexed, but the day isn’t. Which … fine, you just do the math in your head, but it’s annoying
Ultimately this is how I ended up storing the dates:
{
title: "The Dearly Beloved",
display_title: "The Dearly Beloved",
author: "Cara Wall",
cover_image: "the-dearly-beloved.jpg",
genre: "Fiction",
start_date: "2019/9/14 1:00",
end_date: "2019/9/14 23:00"
},
Note the hour information — I don’t do this for all of my entries, just the books that I started and finished on the same day. When I ran the first charts I noticed that the same-day books wouldn’t get rendered at all, so I had these weird gaps in my chart.
I put out a quick query to the Semiotic Slack channel, and Elijah responded:
Oh, interesting. So, technically, if it’s the same day beginning and end then that would be a zero, so I don’t think we’d want to render those. However, you could set the start date with the time equal to 12:01AM and the end date with 11:59PM and that should give you a shape.
I added hours and the gaps went away. While I was fixing that I decided that I should have the data array be valid JSON — which meant that I had to convert the dates before feeding them into the chart, so this is what I do in the rAccessor
property:
rAccessor: function (e) {
return [new Date(e.start_date), new Date(e.end_date)];
},
Formatting the X Axis #
When I first plugged in my data I put in expanded date information and … yikes. It was clear I needed to deal with axis labels:
A little bit of digging revealed that D3 has some handy things: time scales, and time formats. Since Semiotic is pretty much glueing React and D3 together, it seemed like a good next step to figure out how to get time scales. Dozens of tabs later, this is where I ended up. Semiotic has an rscaleType
property which can receive a time scale domain, which is the full set of values for the time scale. Here I’m restricting the time scale to a single year:
const timescale = scaleTime().domain([
new Date(2019, 0, 1),
new Date(2020, 0, 1)
]);
…
rScaleType: timescale
You can also set a timescale format for the tick mark labels, like “Jun 01”
var timescaleFormat = timeFormat("%b %e");
and pass that in to the axes
property, along with the number of tick marks that we want (in this case 12, one per month):
axes: [
{ orient: "top", ticks: 12, tickFormat: timescaleFormat, label: "Month" }
]
Tooltip Annotations #
Semiotic has two properties if you want to do tooltip annotations, pieceHoverAnnotation
and hoverAnnotation
. Set both to true and you have tooltips that are placed relative to your mouse position.
In my case I also wanted to customize the content that showed up within the tooltip, so I used the tooltipContent
property which lets you feed in JSX:
tooltipContent: (d) => (
<div className="tooltip-content">
<div className="tooltip-book-image">
<img
src={require(`./img/${d.cover_image}`)}
alt="Cover for {d.title}"
></img>
</div>
<div className="tooltip-book-meta">
<h2>{d.title}</h2>
<p>by {d.author}</p>
<p>Started: {new Date(d.start_date).toLocaleDateString("en-US")}</p>
<p>Finished: {new Date(d.end_date).toLocaleDateString("en-US")}</p>
</div>
</div>
),
You’ll note that I have to do the same String-to-Date conversion here for the start and end dates.
A Detour into Web Scraping #
Getting book images #
Something I should have done over the course of 2019 was download book cover images. Faced with 120+ items I had to decide whether I wanted to spend a morning downloading these images manually, or dust off my Python skills, read up on a bit of web scraping, and (hopefully) be done in a few hours.
I decided that even if it took longer to automate things, it was a skill that I wanted to learn. I switched to NodeJS from Python early on — I knew I could do it in Python but I wanted the challenge of staying in JavaScript land. My colleague Chris Arasin sent me this tutorial on web scraping with NodeJS, and I was off to the races. Here’s the high-level overview of what I ended up doing:
- Generate the URLs I would need for (gently) scraping my source. This post helped with reading/writing to a JSON file
- Scrape a single book URL, parse the data, grab the image URL, request that image, and save it to the filesystem
- Limit my outbound requests so I didn’t flood the server — I didn’t want to be rude, and I didn’t want to get my IP blocked. The Bluebird library has a promise concurrency option in its
map()
method that basically queues up actions if fed an array, and that meant that it would only request one image at a time
Promise.map(urls, getImage, { concurrency: 1 });
- After all the images were downloaded I cleaned up the filenames: I used a quick Automator workflow to make them all lowercase, found false matches, and manually downloaded the dozen-ish covers for books/comics that weren’t found during my scraping run
When that was all done I was left with a nice tidy set of data including images that I could feed to the Semiotic chart.
Styling the Chart #
Almost home! I didn’t like the default styling for the tick marks and chart axis lines, but a quick query on the Semiotic Slack channel revealed that the axes and tick marks can be styled with CSS. I ended up doing this:
.axis path {
stroke: lightgrey;
fill: lightgrey;
}
.axis-baseline {
stroke: lightgrey;
}
What’s next #
Here’s the result after all of this. I’m pretty happy with how this turned out, even if the chart does look a little too much like a Webpagetest waterfall report.
This has been a lot of fun. I’m thinking about other visualizations, now: what about the number of pages? What about the breakdown of women/PoC, or an intersection thereof? Putting all those different charts into their own React components so that they can be display in series (or in a set of switching views) might also be a worthwhile exercise.
Code for my reading log is here, and the (messy) repo for my NodeJS webscraping is here.