Hello! I am new to stan and i wanted to know if there is a way to implemet a stan code for (multi) Classification using Gaussian Process. Thanks

Yeah, I see why you’re asking the question above. I can paste a proper implementation of a Bernoulli likelihood tomorrow (when I’m at a computer). But I believe for multiclass we can use the multinomial distribution (I think probit can be used for many categories but I haven’t fact checked myself). …

hi…this could be very useful for me too. Thanks

Hey all - @fabio @Kon7 So here’s a “binary classification with GP”, or a “Gaussian prior, with a logistic likelihood”. We can write this in math as follows, the full posterior being: \int p(y | X, f) p(f | x) p(x |\theta) With a non-gaussian likelihood functions, we can’t integrate/marginalize …

it throws me an error " No matches for: gp_matern52_cov(vector[], real, real[]) Function gp_matern52_cov not found. " for the multi class we use the multinomial likelihood with the softmax mapping right? also how can we implement the ARD efficiently? Thank you

Thanks for sharing! I am working on a categorical_logit likelihood, but it is not straightforward for me to figure out an implementation.

Thank you! I saw the post but i think i need the specific kernel gp_matern52_cov because it takes as an input a D-dimensional vector regarding the lenghtscale (i need it for ARD) and the other kernels (cov_exp_quad) cannot… i found the implementation here https://github.com/standev/math/tree/devel…

i tried to implement the multiclass case and i ended up in something like this… functions { vector gp_pred_rng(vector[] x_pred, vector y1, vector[] x, real magnitude, real length_scale) { int N = rows(y1); int N_pred = size(x_pred); vector[N_pred] f2; { matrix[N, N] K = 10 + cov_exp_quad(x, magnitu…

@Kon7 @fabio [image] Kon7: gp_matern52_cov(vector, real, real) Yes, just change it to cov_exp_quad . I was using a matern52 on a local library. This will hopefully be available on next release. cov_exp_quad is the only covariance function in Stan’s current release.

thanks…is there another way that i can implement the ARD inside the code for the multi-class case?

Classification with gaussian process

General

fabio February 27, 2019, 1:49pm 9

Have a look at this post. The error is similar, the solution is the same!

Topic		Replies	Views
Applied Gaussian Processes in Stan, Part 1. A Case Study Modeling	12	2937	November 21, 2019
Gaussian process regression General	40	4078	September 19, 2021
User-defined models for gaussian processes with non-gaussian likelihoods (and other generic models) Modeling specification , gaussian-process	19	1497	May 9, 2018
Using 2D Gaussian process predictions within model Modeling gaussian-process	11	2717	February 18, 2022
Adding Gaussian Process Covariance Functions Developers gaussian-process	81	8865	May 21, 2018

Classification with gaussian process

Related topics