Hello, I modelled my 2-dimensional data (one being a continuous value and one being the category the datapoint stems from) with gaussian processes in 2 ways that should actually be equivalent but somehow I yield different prediction results (the posterior seems fine). The only difference is basical…

Maybe, I wouldn’t be surprised. Cloning a PyMC model is not trivial and I don’t expect deepcopy to do a good job. This is specially relevant for GPs, as they hold references to old variables in the instances. PyMC has a clone_model utility, but for similar reasons that won’t work properly with GP o…

Equivalent GP models, different results

Questions v5

jan5 October 17, 2023, 2:54pm 5

It seems that deepcopy is indeed the problem. This time I used only one model for both fitting and prediction and got the same results as in the approach with 2 separate models created from scratch.

1 Like

Hitting a weird error to do with RNGs in Scan in a custom function inside a Potential

Topic		Replies	Views
Question about "gp.predict": Gaussian Processes Questions	2	868	September 2, 2021
`conditional` and Deep Gaussian Processes v5 gaussian_process	6	1032	April 17, 2023
Multi-output gaussian processes Questions	13	5543	October 22, 2017
Modeling longitudinal data among patients by coupling gaussian processes with random effects Questions	5	1708	September 2, 2021
Shape mismatch when running predictions with Gaussian Processes version agnostic gaussian_process	5	104	March 12, 2025

Equivalent GP models, different results

Related topics