Björn Wehlin: Fast and easy-to-use TDA computations for criminology and beyond

Abstract.

I will introduce my open-source software ‘masspcf’ that makes many computations in Topological Data Analysis (TDA) both fast and easy to use. Masspcf handles data whose size varies per observation, which may include piecewise constant functions, barcodes, point clouds, and other similar objects. A core efficiency feature of the software package is parallel backends, both for CPU and (multi-)GPU. As a use case and motivating example, I will talk about a joint project on criminology with Patricia Brantingham (Simon Fraser University), Vania Ceccato (KTH), Wojciech Chachólski (KTH), and Ioannis Ioannidis (KTH). In this project, we use an optimization procedure over the space of one-dimensional probability distributions to find a topological “lens” under which to correlate non-criminological (e.g., socioeconomic or land use) data with crime data in Stockholm municipality. However, masspcf can be used in many more contexts, both within TDA and elsewhere. The framework that is general enough to handle representations ranging from central bank interest rates to lattice paths. Masspcf is available for Python and C++.