This project built upon gboeing's lovely location-history project from a few years ago.
The purpose is to compute border-crossings, as needed for annoying beurocratic things like taxes and visas.
E.g. for my UK passport application I needed to report all my comings and goings from UK for the past 5 years. I decided spending 10 hours messing with this was much more fun than spending 1 hour working it out by hand from emails and photos :)
Click here to download your Google location history as a JSON file called Records.json
It should look something like this:
{
"locations": [{
"latitudeE7": 337625316,
"longitudeE7": -843593197,
"accuracy": 40,
"source": "sensor_other",
"timestamp": "2011-04-23T18:22:13.086Z"
},
...Run the relevant cells. It should run top-to-bottom if you have all dependencies installed.
This was developed and tested on a 2021 M1 MBP with 32 GB of ram.
The main hurdle I ran into was clustering the GPS data. The example I started from used DBSCAN which always OOMed and killed my kernel. I found some threads about fixing it, but decided to go with HDBSCAN which is a hierarchical extension of DBSCAN that I hoped was more scalable.
In my experience it still choked after about 900k datapoints, but YMMV.