`pgmuvi`: Quick and easy Gaussian Process Regression for multi-wavelength astronomical timeseries

Abstract

Time-domain observations are increasingly important in astronomy, and are often the only way to study certain objects. The volume of time-series data is increasing dramatically as new surveys come online - for example, the Vera Rubin Observatory will produce 15 terabytes of data per night, and its Legacy Survey of Space and Time (LSST) is expected to produce five-year lightcurves for $>10^7$ sources, each consisting of 5 photometric bands. Historically, astronomers have worked with Fourier-based techniques such as the Lomb-Scargle periodogram or information-theoretic approaches; however, in recent years Bayesian and data-driven approaches such as Gaussian Process Regression (GPR) have gained traction. However, the computational complexity and steep learning curve of GPR has limited its adoption. `pgmuvi` makes GPR of multi-band timeseries accessible to astronomers by building on cutting-edge open-source machine-learning libraries, and hence `pgmuvi` retains the speed and flexibility of GPR while being easy to use. It provides easy access to GPU acceleration and Bayesian inference of the hyperparameters (e.g. the periods), and is able to scale to large datasets.