How to Build Interactive Choropleth Maps with Python, Vega, and GeoJSON

When I started building Vincent, my goal was to make map creation as rational as possible. There are great Python mapping libraries—Basemap and Kartograph—that make maps more interesting, and I strongly recommend them because they are powerful and easy to use.

I wanted a simpler tool that leverages Vega, allowing a concise syntax to point to a GeoJSON file, describe a projection and size/scale, and finally output a map.

For example, layering map data can create more complex maps:

Choropleth maps need to bind Pandas data, mapping data columns directly to map features. Assuming a 1:1 mapping from GeoJSON to column data, the syntax is very simple:

Our data must have a 1:1 mapping between GeoJSON keys and the Pandas DataFrame. The required CSV contains a column with FIPS codes, country names, and economic information (column names omitted):

In the GeoJSON, country shapes use the FIPS code as the id (thanks to a fork of Trifacta). Simplified shapes are shown here; the full dataset is available in the example data:

We need to match FIPS codes correctly, otherwise Vega cannot properly compress the data:

Now we can quickly generate different choropleths:

The maps show that Los Angeles and King counties are very large and densely populated. Next we look at median household income, revealing many high‑income areas on the East Coast and other dense regions:

Finally, we reset the map to view national unemployment rates:

Mapping is my passion—I hope Vincent becomes stronger, supporting easy addition of points, markers, and other capabilities.