Weighted K-Means algorithm for Redistricting in North Carolina

Developing an unsupervised ML method for drawing the voting districts in North Carolina.

Algorithm

The original Weighted K-Means algorithm has been proposed by Guest et al. (2019) and the original paper can be found here: https://link.springer.com/article/10.1007/s42001-019-00053-9

I extend the method by weighting the distances not only by population but also by racial and partisan differences. Moreover, I open-source my code in hopes to continue the conversation about computational redistricting.

Data

1. centroids.csv: Longitude and latitude for each Voting Tabulation District (VTD), the most granular data I could get

2. demo_NC_VTD.cpg, demo_NC_VTD.dbf, demo_NC_VTD.prj, demo_NC_VTD.shp and demo_NC_VTD.shx: VTD level population, racial and voting data as well as shapes of each VTD for plotting. centroids.csv is a derivative of these shapes, so the order of rows is the same.

3. demo_SSP_asrc.csv: Projected zip-code level population data from https://osf.io/uh5sj/ subsetted for North Carolina in 2030 and 2050

Demonstration

weighted_kmeans.py is a Python script that implements the Weighted K-means from scratch.

A demonstration of how to use the method can be found in demo_weighted_kmeans.ipynb. The notebook also includes visualizations of the proposed maps for 2016 as well as the projected maps in 2030 and 2050 based on this paper https://www.nature.com/articles/sdata20195.