search

dfm / george

Fast and flexible Gaussian Process regression in Python
http://george.readthedocs.io
MIT License
451 stars 128 forks source link

Uncertainties blow up #166

Closed temuller closed 9 months ago

temuller commented 9 months ago

Hi, I am trying to fit some supernova light curves in 2D (flux as a function of wavelength and time) and george works nicely, except that the uncertainties blow up wherever there is no data. I have tried different constrains on the hyperparameters, but without good results. I am using a matern 5/2 kernel on the time axis and squared exponential on the wavelength axis.

See below an image with an extract of a single light curve from the 2D fit. Any ideas on how I can improve this? As a reference, my code can be found here: https://github.com/temuller/piscola/blob/master/src/piscola/gaussian_process.py#L92

Screenshot from 2024-02-05 16-50-44

dfm commented 9 months ago

@temuller — This actually looks as expected to me! A GP isn't really a good model for non-stationary signals like these, but that's more of a research question than a george issue. Please feel free to email me if you'd like to chat about it!