In a previous post I discussed how to reduce the size of a spatial data set by clustering. Too many data points in a visualization can overwhelm the user and bog down on-the-fly client-side map rendering (for example, with a javascript tool like Leaflet). So, I used the DBSCAN clustering algorithm to reduce my data set from 1,759 rows to 158 spatially-representative points. This series of posts discusses this data set in depth.
Read/cite the paper here.
In this tutorial, I demonstrate how to reduce the size of a spatial data set of GPS latitude-longitude coordinates using Python and its scikit-learn implementation of the DBSCAN clustering algorithm. All my code is in this IPython notebook in this GitHub repo, where you can also find the data.
Traditionally it’s been a problem that researchers did not have enough spatial data to answer useful questions or build compelling visualizations. Today, however, the problem is often that we have too much data. Too many scattered points on a map can overwhelm a viewer looking for a simple narrative. Furthermore, rendering a JavaScript web map (like Leaflet) with millions of data points on a mobile device can swamp the processor and be unresponsive.