Trying to predict new values using .conditional of a Latent GP using my own 2D kernel definition results in a incompatible shape error (MWE at the end). In a nutshell, I’m creating a 2 dimensional (spatial) Gaussian process model using Haversine instead of Euclidian distance. My kernel definition i…

The problem is incorrectly implemented broadcasting within the great circle distance which should be computed as follow: def great_circle_distance(self, X, Xs=None): if Xs is None: Xs = X # Assume first column is longitude and second is latitude lat1_ = pt.d…

Prediction with Latent GP using 2D kernel definition

Questions v5

nrieger June 28, 2024, 9:38pm 3

Please note: In general, Haversine distance + Matérn kernel is not a valid combination. Please check the discussion here.

1 Like

Topic		Replies	Views
Haversine Kernel? v5 gaussian_process , modeling	6	133	June 26, 2024
Multivariate Gaussian Processes Questions	2	526	July 27, 2020
GP.conditional Input dimension mis-match Questions	3	509	February 15, 2021
The predict function of GP Questions	10	1192	June 15, 2021
Correctly Defining Gaussian Process Polynomial Kernel Questions	0	461	July 14, 2019

Prediction with Latent GP using 2D kernel definition

Related topics