Trends — Hourly Items

February 3, 2020

Share on FacebookShare on TwitterShare on LinkedIn

Trends — Hourly Items

Click here for more info on this project.

24 hours

Predicting when any one person will want a coffee is hard. Most of the time it will probably be in the morning, there might be the occasional afternoon coffee break, etc. But millions of people’s habits over three years form a very clear pattern. In this case we’re hearing what time of the day people are ordering three items: coffee, tequila, and condoms.

The way I’ve always thought of data like this is that this is the probability that any one of us will order a coffee at a given point in the day. It just manifests itself in a very clear form when we aggregate everyone’s activity together.

The Items

Maybe you didn’t know this, but you can get almost anything on Postmates, from tacos to toilet paper. You might be used to getting burritos or burgers, but if there’s a local merchant that sells something, we can get it for you.

That’s part of the reason why I chose these items. Everyone orders food in a pretty similar pattern. Of course there are differences, but they’re subtle, people generally get hungry in a predictable fashion.

Once we start looking at non-meal related items, we start to see behavioral patterns for things besides hunger. Getting a caffeine boost happens in the morning, winding down with a drink happens mostly at night, and getting intimate peaks later at night. Interestingly, condoms don’t have as much of an off-time as the other two. That kind of makes sense, not many people are in the mood for tequila at 6am, or coffee at 1am, but people can be in the mood just about any time.

What You’re Hearing

I’m using the same loop of random sampled instruments with different starting points and notes for each category. I’m using a sample library from one of the most unique virtual instrument creators out there, Sound Dust. The volume of each loop is determined by the distribution of deliveries in 10-minute intervals throughout a day.

It follows the chart above, we start at midnight, and condoms are the most prominent, then we hear a quick, loud spike for coffee in the morning, then we hear tequila start to swell, and condoms again late at night to bring us back to where we started.

Because the loops are so long and start at different points, you won’t really be able to hear any repeating patterns in the notes themselves. And that’s the point, to just create a texture where the change in volume is the most prominent characteristic, which is controlled by the data.

Here’s what each track sounds like on its own:

How it Works

I made a simplified tutorial on how to create these from start to finish. Check out the Colab notebook if you want to make one yourself.

This one was relatively straightforward in terms of creating the three MIDI tracks. The curves you saw in the chart above were scaled to integers between 0–127 and assigned to a MIDI CC track, each series is assigned a note that is sustained through the whole piece, and that CC track controls the volume of each note. That sample notebook is all you need to recreate this one.


The point with this one is pretty simple, but it’s a great one to highlight how interesting sonification can be. People want coffee in the morning, tequila at night, and condoms late at night, that’s not going to win any nobel prizes. But because not all people behave identically, you can hear the ebbing and flowing of people’s changing desires throughout a day.

Music is a great way to demonstrate this. You can just look at a chart and then look away. If we turn these slow transitions into something that has to be experienced, we can really feel the desire to get one particular item diminish as the desire to get another grows.

Postmates is always looking for creative data-focused people to join our team. If you want to make things like this, check out and say that Alex sent you.


More from Engineering

View All

Density — Sound Terrain

I’m sure you already know this, but The United States of America is a huge place. And I’m sure you also know that the population is not evenly distributed. So that’s not interesting.

February 3, 2020