Thomas J. Fan - Time Series EDA with STUMPY (April 10, 2024)

STUMPY is a robust and scalable Python library for computing a matrix profile, which can create valuable insights about our time series. STUMPY is built with Numba, which parallelizes computing on CPUs and accelerates it with GPUs. In this talk, we learn about the matrix profile and the various methods to compute it. Afterwards, we explore how to use the matrix profile in applications such as pattern discovery, anomaly detection, semantic segmentation, and time series chains. Thomas J. Fan is a senior machine learning engineer at Union.ai and a maintainer for scikit-learn. At scikit-learn, he led the development of DataFrame interoperability and GPU support through PyTorch. Previously, as a researcher at Columbia University, Thomas collaborated with NASA to automate machine learning workflows.