Tutorial and python implementation for carrying out Stochastic Variational Gaussian Process Regression on large datsets with modelling for input-dependent noise profiles.
This method allows one to simultaneously infer both the underlying latent function and input-dependent noise profile of large 1D datasets. Originally developed for applications in data-driven Galactic Dynamics but is applicable to any dataset with input-dependent heteroskedastic noise.
Based on the method presented in "Gaussian Process Methods for Very Large Astrometric Data Sets (Hapitas et al. https://arxiv.org/abs/2507.10317, awaiting publication in ApJ).
- Under development - full tutorials and code will be added following paper publication. *