07-03, 11:00–11:25 (Europe/Amsterdam), Momentum 2
Language: English
Overview: wSTKDE is an open-source Python/Numba project for high-performance spatio-temporal kernel density estimation.
Performance: Implements a custom algorithm that scales linearly with the number of input points, and processes a real-world dataset of 1.7 million points in just 0.5 seconds.
Background: Developed in part during a master's thesis in Geo-Information Science.
Availability: Released under MIT license on GitHub.
Spatio-temporal kernel density estimation (STKDE) is a method for identifying and analysing patterns of points over space and time. This method is commonly applied in fields such as epidemiology, criminology, transportation, and environmental science to detect hotspots and monitor trends.
In this talk, I’ll start with the fundamentals and use cases of STKDE, using a bit of math and some animations to illustrate how 1D KDE works. We’ll then look at the limitations of existing implementations before diving into the key performance optimizations behind wSTKDE’s custom, linear-scaling algorithm. Finally, you can expect a demo showing how to process 1.7 million NYC traffic accidents in under half a second.
In short, this talk will be of interest if you have a use case that could benefit from this algorithm, are curious about the math behind kernel density estimation, or just enjoy well-optimised algorithms.
MSc in Geo-Information Science.