I don’t really love running. I run marathons. These two things are true at the same time. I started running a lot (like, a lot) during the lockdown 2020 and somehow never stopped. While I do enjoy the actual act of running, the one thing that keeps me coming back to it is the music. My playlists. To me, running is just a different form of dancing – moving your body to music.
As I got more into training for actual races, I noticed that running to music at a certain bpm helped me keep up my cadence and made running a goal pace easier. Running cadence is basically the same as bpm for music, it describes how many steps you take per minute. A higher cadence is generally considered more favorable as it reduces the time spent on the ground, and quicker turnover can help you run faster. I generally aim for 175 bpm for an easy/medium run which gets me to a 9:30 pace with light effort, and faster bpm for faster runs.
You probably know where this is going by now: I’m always looking for a good playlist with my target tempo. There are obviously a bunch on Spotify already, but they’re not necessarily the songs I like. While Spotify recently introduced auto-generated playlists at a certain bpm based on your liked artists, it seems to have a pretty wide margin for the tempo. Since I’ve been wanting to play around with auto-generated code by ChatGPT for a while, I figured I might as well tell the AI to write some code for me. And it went… ok?
Here’s my first prompt, just to test the waters:
>> Can you write me some python code that gets a list of all my liked songs including their ID and bpm from Spotify?
Sure, I can help you with that. To get a list of all your liked songs including their ID and bpm from Spotify, you will need to use the Spotify API and the spotipy library for Python. Here is an example code that should work:
You will need to set up a Spotify developer account and create a new app to get the required credentials for the Spotify API. Then, replace the client_id, client_secret, and redirect_uri values in the SpotifyOAuth call with your own credentials. Finally, run the code and it should print a list of your liked songs with their IDs and BPMs.
That already looked pretty good, so I got a bit bolder and asked ChatGPT to actually give me the full code to add songs based on a certain bpm… and it delivered:
This looked like reasonable code to me. I was particularly surprised and somewhat bewildered that it automatically added the 5 bpm margin. Of course, most human developers would know that a buffer makes sense and ChatGPT just learned from existing code, but it almost felt like it was being smarter than what I asked it to do?
However, when I tried to run the code, I encountered two minor problems:
First, the example didn’t actually add the variables for the authentication credentials in the SpotifyOAuth call that it mentioned in the comment after the code. I had to do some googling to figure out how to set use these and decided to set the respective environment variables instead.
And second, the data model for the object returned by the Spotify API has changed, and the “tempo” now requires a separate API call. I found an example API call and asked ChatGPT explicitly to use the different method to get the “audio features” object:
>> can you write some code that uses the spotipy library to retrieve the audio features for a given track ID
I’ll skip the code snippet it gave me, but I ran the code separately and it ran just fine. Then I put everything together and it… just worked. Playlist created. Job done.
The final result
Overall, this took me maybe 15 minutes to put together, including figuring out authentication (I had already registered for a developer account before). Even though the code is fairly trivial, the biggest help was the API examples and the ability to ask questions about the API. It felt like working with a buddy who just happens to know that particular library and API pretty well and can you point the right way. I actually think I might use this in the future when working with unknown APIs. I’m decent enough at reading library and API docs, but using ChatGPT just really sped things up, even though it initially pointed me to an outdated version of the data model.
I don’t actually know how this would work for inexperienced developers that don’t have a good mental model of how the code should be structured at all and might have struggle with debugging error messages if the code doesn’t actually run. But I feel like for someone who’s written enough code to have an outline in their head and just needs some examples for things like API calls, this might be a great way to accelerate writing code.
The notebook with the final version of the code can be found on my GitHub. And yes, I did run the London marathon with (part of) that playlist!
Sam Bail 
You must be logged in to post a comment.