Flourish – JazzGraphs

Tag: Flourish

Visualizing Musical Genres, Part 2

Posted on March 20, 2024 by jazzman

My last post showed a little bit about the musical genre data structure in the MusicBrainz database. In this post we’ll expand our view to include all genres and sub-genres, and look at a few visualization approaches using Flourish.

Flourish provides several options for visualizing hierarchical data; in this post we’ll look at some of the advantages and disadvantages of each approach. Ultimately, my goal is to categorize all my CDs and vinyl using this approach, but for now we’ll work with the MusicBrainz genre data.

We had a quick look at a sunburst chart in the prior post, so we’ll begin there with the much larger dataset we now have. How does it work?

Sunburst chart displaying all genres and sub-genres

Hmmm…it’s a little challenging to see the data beyond the first few genres (these are the ones with the most sub-genres). We can narrow our focus by using the filter or by clicking on one of the inner circle genres. Let’s look at the rockgenre:

That’s a bit better; note that each sub-genre has an identical size here, something that will change once I feed my own music collection into Flourish. At least we can now identify all the sub-genres in the data.

What about a treemap approach? Treemaps can be useful in showing categories and sub-categories, sized by count or some other value (revenue, sales, profit, etc.). Here’s a look at all the data:

Once again, it’s a challenge to see anything beyond the most frequently occurring genres; even if we provide a pop-up label it’s not very user-friendly. Let’s filter down, this time in the electronicgenre:

Here we get a similar result to the sunburst, albeit in a different layout. Again, this could be more interesting with an actual record collection, where each sub-genre would potentially be sized differently, with some not even appearing (i.e.- no recordings in a sub-genre).

Our next example will use circles, an approach sometimes known as circular packing. All genres will be arranged in a somewhat random layout, rather than the radial or rectangular formats we have just seen. Here is a look at all genres:

Once more, we have a similar issue to the sunburst and treemap displays, although it is fairly easy to see the highest frequency genres in the center. Filtering on the popgenre yields a series of identical sized circles for all pop sub-genres:

The circle approach is perhaps my least favorite of the three we have seen thus far, due to the seemingly more random placement of the individual circles.

At the opposite end of the spectrum we can use bars to view the same data. Here we are able to clearly see the rank order and relative frequency for each genre:

This looks really good for the high frequency genres – clear labels with easy to distinguish relative frequencies. The downside is when we have hundreds of genres; our bar chart becomes incredibly tall from top to bottom. In short, this approach will be effective for a limited number of genres, although the same could be said for the other methods.

Our final approach uses a radial tree option in Flourish. This method most closely mimics the sunburst option, with results laid out in a circle; genres can then be clicked on or filtered to get to the sub-genre level. Here are all genres:

Not exactly helpful, is it? There are simply too many genres and sub-genres to display; even the sunburst chart provided more information at first glance. But what about when we select a single genre, such as reggae?

That’s better! We now have a clear, concise display to work with. This could prove to be useful when we have different size values for each sub-genre; in essence it will merge the best aspects of the sunburst and bar displays. I’ll be interested in seeing this sort of display when my music collection data is complete to see how well it handles differing sizes.

So which approach is best? I’m going to say that it depends on the underlying data; none of these charts was great when we attempted to view all genres at once, but they do appear to offer potential when the data has fewer categories (genres). Personally I like the sunburst and radial methods for the clarity of their display coupled with the visible connection between the sub-genres and the parent genre. I’m eager to see how they work with a more typical dataset.

That’s it for now – hope you enjoyed this, and thanks for reading!

Musical Genres via MusicBrainz Data

Posted on March 14, 2024March 14, 2024 by jazzman

Once again I have pulled the core MusicBrainz tables into my local version of PostgreSQL, where I can start exploring all sorts of musical data – recordings, releases, places, artists, and much, much more. The database is large, totaling nearly 17 gigabytes of data across 171 tables, so there is no shortage of potential topics to explore.

One of the areas that intrigues me the most is an exploration of musical genres, created in MusicBrainz for contributors to categorize recordings. While the genres don’t currently tie in to individual recordings or releases in the database, I hope to use them with my own collection of music to create some potentially interesting visualizations. For now, let’s undertake an exploration of the raw data on genres, using the DBeaver database tool.

Our first table is simply named genre; let’s look at a screenshot of some of the data:

Each genre has a unique id and gid value linked to a distinct genre name such as acid house, arena rock, or bebop. As you can tell, we’re going to a very specific level here, not simple classifications like pop, rock, or jazz. This should make it quite interesting (but not so easy!) when I start tagging my own music collection.

A second table is named genre_alias; here we find some examples where distinct names are rolled up to a single genre id to join to the genre table we just saw. For instance, have a look at some of the entries below:

Data from the MusicBrainz genre_alias table

We see multiple rows pointing to a single genre id (the genre column), largely based on alternative spellings or differing punctuation. The last three rows display one such case – alternative rap, alternative hip-hop, and alternative hiphop all have a genre value of 10; in the genre table this classifies all three as alternative hip hop. In other words, these entries are three possible variations on the original alternative hip hop genre; they all represent the same musical genre. In a sense, this is some data cleansing that I won’t need to perform.

A third table is named l_genre_genre; it ties together sub-genres with a higher level ‘umbrella’ genre. Using the alternative hip hop example from above, let’s dive into this table, where we can see the 10 value in the entity1field:

Alternative hip hop id in the entity1 column

Note the 199 value in the entity0 column; if we refer back to the genre table, here’s what we see:

The 199 id value corresponds to hip hop, which contains the alternative hip hop genre, as well as any other sub-genres related to hip hop. So entity0represents the higher level grouping, with entity1 representing the next level down (a sub-genre). In terms of classifying music, we can now use two levels, which may prove useful when it comes to building visualizations.

Is there anything we can visualize at this early stage? How about a very simple sunburst chart? These will become far more interesting when I can tag my own collection with genre info, but for now, here’s a conceptual look using Flourish. You can use the filter or click on a genre to focus the display.

I hope you can see the potential here; ultimately each genre and sub-genre will be sized based on the number of albums (vinyl & CD) in a collection. This will provide a quick visual indication for where someone’s musical preferences lie. There may even be the possibility to take it down to the single recording level, but we’ll have to test that idea.

That’s it for now; looking forward to doing some more fun stuff with the MusicBrainz data. Thanks for reading!

Miles Davis Song Plots

Posted on January 12, 2023February 18, 2024 by jazzman

In this blog we’re going to use Flourish with more MusicBrainz data to plot the length of Miles Davis songs on a range of vinyl releases. This type of data often suggests the use of a scatter plot with an x-y axis to best visualize the information. For instance, we could place record labels on the x-axis, and the length of each song (in seconds) on the y-axis. However, with record labels being a categorical variable (i.e.- discrete values such as Sony, Columbia, etc.) there are better options for understanding the data versus a true scatter plot.

The first of these is a boxplot, which provides the ability to see the distribution of data (song lengths) by record label. Let’s take a look at this data in Flourish:

Here we have limited the data display to a single label (showing all was quite messy!). Select CBS or Columbia to see labels with many Miles Davis releases. We now see the median length of a recording, as well as the 25th percentile (bottom of the box) and the 75th percentile (top of the box). It’s also easy to see individual songs that lie below or above the typical range; in statistical terms, these are called outliers. On our plot, they represent songs that are either much shorter than normal (below the extended line) or longer than normal (above the extended line).

This is all useful information, but presents some limitations. Boxplots are very good at doing the aggregations for us while obscuring the individual data values, especially values that lie inside the box. To improve our ability to see those values we turn to a violin plot, which excels at showing the shape of a distribution, rather than the fixed shape provided by the boxplot. We have also combined a beeswarm plot with the violin plot so we can see every individual value:

Again, select CBS or Columbia to view a label with many releases/songs to understand why we elected to use this approach. Hover over individual points to learn more about an individual song – it’s length, release, artist, label, and song title. For me, this approach is best if I’m trying to explore the data; the boxplot is great when I’m interested in overall patterns. Both are powerful tools suited to their individual strengths.

I’ll be using Flourish to interrogate the MusicBrainz data further in future posts, but that’s it for now. Thanks for reading!

Miles Davis Sunburst Visualization

Posted on December 26, 2022February 5, 2024 by jazzman

With the Christmas holiday chaos (somewhat literally this year) in the rearview, I’ve been playing a bit with the MusicBrainz data and the Flourish visualization library. First up was using some repurposed code to visualize Miles Davis recordings. I thought a sunburst diagram might be an interesting way to show album releases and the songs on each release. Turns out it wasn’t quite as simple as I thought…it never is!

After multiple query tweaks and iterations, I’ve got something fun and interesting. Miles produced so much music, with much of it re-released in multiple formats (think vinyl vs. cd) and in various collections, factors that wound up influencing my query and chart logic. As is the case for many jazz artists, multiple labels are an issue, so why not create a filter to view releases for each label (Columbia, Blue Note, etc.)? And many songs turn up on multiple releases (studio, live, collections), so we need to account for that as well.

So my thought with using a sunburst was to group songs and releases together, and allow filtering by label. Mind you, it took multiple attempts to get the data in the best format, but we eventually wound up with something workable to feed the sunburst chart.

If you aren’t familiar with the sunburst chart, here’s a quick primer. The goal of a sunburst chart is to display hierarchical information in a circular layout with 2 or 3 levels (typically). The outer layer has more surface area to work with, and successive inner layers each have less visual space to use. For this reason, I wound up using individual songs in the outermost layer, with their respective albums as the inner layer. With an average of perhaps 5-10 songs per album, this takes advantage of the sunburst hierarchy framework.

Here’s what the code eventually became, after multiple iterations:

SELECT distinct ac.name AS artist, l.label_code, l.name AS label_name, r.name AS release, mf.name AS format, t.name AS id, t.name AS label, t.name AS name,
r.name AS recording,
CASE WHEN t.length < 180000 THEN ‘< 3 Minutes’ WHEN t.length < 300000 THEN ‘3-5 Minutes’ WHEN t.length < 420000 THEN ‘5-7 Minutes’ WHEN t.length < 600000 THEN ‘7-10 Minutes’ WHEN t.length > 600000 THEN ’10+ Minutes’
ELSE ‘No Length’ END category

FROM public.release r
INNER JOIN public.artist_credit ac
ON r.artist_credit = ac.id
INNER JOIN public.medium m
ON r.id = m.release
INNER JOIN public.medium_format mf
ON m.format = mf.id
INNER JOIN public.release_label rl
ON r.id = rl.release
INNER JOIN public.label l
ON rl.label = l.id
INNER JOIN public.track t
ON m.id = t.medium
INNER JOIN public.recording re
ON t.recording = re.id

WHERE r.artist_credit = 1954
and mf.name = ’12” Vinyl’

ORDER BY l.name, r.name

What we’re doing here, in a nutshell, is retrieving all the information for Miles Davis’ 12″ vinyl releases; many of these recordings were eventually released on CD, so we’re attempting to avoid duplication here. The ‘r.artist_credit = 1954’ line refers to Miles Davis and his MusicBrainz artist ID, while the medium_format name field is set to grab just 12″ vinyl releases.

Enough of the technical details – let’s view some results:

Here’s a look at the dropdown filter we created using labels:

Note that we ordered our query by both label name and release name; this translates to an alpha sorted dropdown on labels, making it much more intuitive to select a specific label. We can choose to display all labels, but that gets rather messy for an artist like Miles who recorded for or was re-released by many companies. Let’s filter it down to Columbia, a major label who Miles recorded for many times:

The inner circle displays individual releases, of which there are many, while the outer ring displays the songs on each release. The Flourish sunburst charts are interactive, but it’s a challenge to see what’s going on in our static image. Let’s move to the Blue Note label, a major force in jazz, but one where Miles was not a major player:

Now we can see the layout, with album releases surrounded by individual songs. We can go a step further by clicking on the Miles Davis, Volume 1 layer, which reveals the following:

Now we are focused strictly on that release and can easily view the songs on that album. Hope you get the general idea for how the sunburst charts work. Now have a go at it yourself with the live version:

I’ll have more of these to come, as it feels like a great way to capture a lot of information in a fun, interactive layout. See you soon, and thanks for reading!

Mapping Venues with Flourish

Posted on December 18, 2022February 5, 2024 by jazzman

I recently shared some venue maps created in Carto using MusicBrainz data where we can see all the musical venues (with lat/lng coordinates) listed in the database. Now that I am exploring Flourish as a visualization tool, it is time to test the mapping capabilities to see if it provides a viable alternative for geographic visualization.

As with the Carto map, the first step is to retrieve the data with a query from my local MusicBrainz database:

select p.”name” , p.address , p.coordinates , split_part(trim(p.coordinates ::text, ‘()’), ‘,’, 1)::float AS lat,
split_part(trim(p.coordinates ::text, ‘()’), ‘,’, 2)::float AS lng,
pt.”name” as type, ar.name as locale
from place p
inner join place_type pt
on p.”type” = pt.id
inner join area ar
on p.id = ar.id

where p.coordinates is not null

The results are exported to a .csv file for ingestion by the Flourish map template. In this case we are using a point map to display venues using the latitude (lat) and longitude (lng) data we just created. Before uploading our data, we can see that Flourish provides an easy to use template populated with sample data, allowing us to see the data format we need to deliver:

Here we can see a name field (City) and latitude and longitude fields for positioning points on a map. Other attributes provide data that might be used for sizing, coloring, or context.

In our case, we can upload our venues .csv file in this general format, and then tell Flourish which columns to use. Here’s what our data looks like after uploading:

Now that the data has been uploaded we can specify which columns to use. Rather than using column names (Name, Address, etc.), Flourish works off the actual column locations (A, B, C, etc.). These are the values we need to input in the options tab. We can use A, B, and F for describing the data with labels and/or popups, use D for latitude and E for longitude, and switch from the Date to the Preview pane.

When we complete that task, Flourish gives us an attractive map, with many templates to choose from, and dozens of options for styling the map and surrounding area:

Here’s a quick view of the many available categories for styling your map:

Template options for mapping with Flourish

We’ll skip the details for now, but it is obvious that we could do an amazing number of things to customize our map.

To answer my earlier question, it appears that Flourish will be more than capable for mapping a variety of data, and I’m looking forward to testing it with much larger data sets. See you soon, and thanks for reading!