I am trying to draw samples from a GP prior with Gibbs covariance - following the example in the API on the website if I just were to replace the tanh_func with a function of my own choosing (in this case I am sampling the warping function from another GP) – it is very clear that the resulting covar…

I wasn’t able to figure out why your example as written didn’t work. It looks like this is using exp(GP) for the lengthscale of another GP. Does this give something sensible for samples from the GP prior?: import numpy as np import pymc3 as pm import matplotlib.pyplot as plt import theano.tensor…

Behaviour of Gibbs kernel seems off

Questions

vr308 July 24, 2021, 5:40pm 4

thank you, the behaviour of np.random.multivariate_normal is a bit worrying because even if I can sample …the fact that the matrix isn’t psd might break something downstream.

1 Like

Topic		Replies	Views
A very simple GP fails, giving not positive-semidefinite covariance and nan prediction Questions	11	2788	September 27, 2018
Non Stationary GP Regression Questions	4	963	July 4, 2018
Multiple (uncertain) function observations of the same Gaussian process Questions	15	6568	October 16, 2017
Multidimensional gaussian process Questions	55	10240	October 22, 2020
Understanding the Predictive Covariance of Gaussian Processes Questions	1	459	June 23, 2018

Behaviour of Gibbs kernel seems off

Related topics