Using clustering to create a new D3.js color scale – Medium
►https://medium.com/@dschnr/using-clustering-to-create-a-new-d3-js-color-scale-dec4ccd639d2
In late July 2015, Tom replaced the Jenks algorithm in simple-statistics with a port of an algorithm called Ckmeans–a 1-dimensional clustering algorithm created by Haizhou Wang and Mingzhou Song that is a slight improvement over Jenks in most cases.
Despite being an improvement over Jenks, the Ckmeans algorithm still had a runtime complexity O(kn²) which would take 300–500ms for visualizations that have thousands of data points such as the US counties choropleth above. This is just too slow to be used reliably in applications where performance is important.
Luckily Wang & Song released Ckmeans 3.4.6 in May 2016, with a new core algorithm that uses a divide & conquer dynamic programming approach to achieve O(kn log(n)) runtime. For web applications this is a huge win–the same US counties choropleth now takes 30–50ms to compute clusters, which is a 10x improvement.
#d3.js #échelles #K-means #CKmeans #Jenks
voir aussi ▻http://mappemonde.mgm.fr/geovisu/119