In stan user’s guide a latent variable GP is defined as: F\sim multivariate\_normal(0,K(X|\alpha,\rho)) Y\sim normal(F,\sigma) In my problem I have data for multiple years: t=0,1,… So my model is: YT_t\sim multivariate\_normal(YT_{t-1},K(X_t|\alpha,\rho)) Y_t\sim normal(YT_t,\sigma) where YT …

[image] linas: I found that it is hard-coded then sampling almost halts It’s because the model isn’t specified correctly. I was being nice earlier, but the model doesn’t make sense. I’ve defined how to use GP in time series data above, with time data being one of the X dimensions. I’m happy …

Thanks, this is why I wanted to get opinion from experts. And I appreciate your help. What is your opinion about GP with errors in variables? Relative year - is the year from when experiment starts: 0-5 DBH - proxy for tree height in inches There are two types of treatment: T and X. DistU - dist…

Thanks I’ll get back with you in a week or so.

[image] linas: While first two are parameters of Kernel the last one is small nugget that is used in some R libraries for fitting with GP. I don’t remember which but I can find out. Then I add observation error to the true value of canopy. If you have a separate term for the observation error…

I meant that parameters in AR(1) may be modeled as GP. Just a thought…

[image] linas: I meant that parameters in AR(1) may be modeled as GP. Just a thought… Why ?

Well, for each tree the change from year to year is governed by covariates that don’t change between years so the \alpha, \beta in AR(1) should be constant year to year. But the change is too random to be explainable just by normal noise. So I thought to model the change as GP. In fact I tried AR(1)…

[image] linas: It is not in cmdstan 2.21.0 either. Really strange. Can you upload your version, in command Stan of ‘/stan/src/stan/lang/function_signatures.h’? Should be there, I’ve added this to the language many months ago. It should be there unless someone’s deleted it.

Hi @linas can you please also tell me if you remember whether you’ve cloned or installed from tar for command Stan?

@linas Earlier, you mentioned that when a trees canopy starts to die, you say that the tree is dying and you stop taking measurements. Do you have a measurement of canopy lost? Do you have a unique identifier for all of the trees? Typically in this case we can use a time varying survival mode…

Time series and Gaussian process

Modeling

Raoul-Kima November 21, 2019, 7:14pm 30

Why ?

rStan math version

Topic		Replies	Views
Gaussian Process with measurement uncertainty and a large number of observations Modeling rstan , fitting-issues	15	346	March 16, 2026
Gaussian Process out-of-sample predictive distribution Modeling gaussian-process	21	2474	December 6, 2024
Binary time series Modeling	27	2287	February 17, 2020
Using 2D Gaussian process predictions within model Modeling gaussian-process	11	2622	February 18, 2022
Modelling phylogenetic and spatial autocorrelations as Gaussian processes with multiple observations per species and quadrat Modeling gaussian-process , spatial , phylogenetic	8	1426	September 9, 2021

Time series and Gaussian process

Related topics