The way I understand them, in a Deep Gaussian Process, the output of each layer becomes the input to the previous layer. How does one predict on new points using these models? Do I need to call gp.conditional on just the input?

I think you’d have to call gp.conditional for each of the GP layers and feed them into each other. I think it’d be super slow though. You could try using pm.gp.HSGP instead of pm.gp.Latent for this.

`conditional` and Deep Gaussian Processes

Questions v5

Thops April 13, 2023, 8:55pm 3

Yeah, I notice HSGP recently but don’t really understand it yet. I’ll have to look into that

Topic		Replies	Views
Predictions from Latent GP in hierarchical model appear not conditioned on anything Questions	6	832	July 6, 2020
Multivariate Gaussian Process Classification (GPC) and Prediction Questions	2	915	February 23, 2020
Using a Gaussian process at intermediate level in hierarchy Questions	3	899	May 17, 2020
HHMM with GPs for state transitions Questions	1	541	July 24, 2018
Predicting on new data with gp.conditional Questions	7	1360	November 10, 2020

`conditional` and Deep Gaussian Processes

Related topics