Hi, I am able to run latent variable Gaussian process model successfully on several data sets (A or B) separately or their combination (A+B). However, when I change variance model from y\sim normal(f,\sigma) to y\sim normal(f,\sigma_1+\sigma_2*y*(1-y)) I am able to run the model on A but not on B. …

This is not working. I wonder how to express your suggestion in mapped function.

I highly recommend first to code a non-parallel version and then transform it to a parallel version. Do you think about using map-rect or reduce-sum?

The problem is that I already have the version with normal coded, see below. It is making good predictions already but I am a perfectionist and looking for even better. Also I don’t like using normal for data in [0, 1]. vector lp_reduce(vector phi, vector theta, real[] xr, int[] xi) { int N = xi[…

Here is the current model. The model predictions for in sample data are quite good but since observed variable is in [0,1] I am not confident that normal distribution is best. Objective. Model canopy of trees over several years. Trees are infected by some bug and are being treated with insecticide …

You can always consider a logit transformation.

Thanks. Problem is that there are bunch of y=0 or 1.

Do you mean logit(\tilde y^x_t)\sim normal(yT^x_t,\sigma) where \tilde y_t^x=y_t^x+\delta and \delta=10^{-5} if y_t^x=0, -10^{-5} if y_t^x=1, 0 otherwise?

If that hack gives good results, than yes, that’s what I mean ;-)

Is there a better way to implement this hack? for (i in in 1:N) y_bar[i] = if_else(y[i] == 0, y[i] + 1e-5, if_else(y[i] == 1, y[i] - 1e-5, y[i]));

Or rather for (i in 1:N) y_bar[i] = y[i] == 0 ? y[i] + 1e-5 : (y[i] == 1 ? y[i] - 1e-5 : y[i]);

Variance model

Modeling

maxbiostat May 30, 2020, 3:12pm 48

You might want to look into LOO. There’s a helpful FAQ as well.

Topic		Replies	Views
Simple model that converges in JAGS but not in Stan Modeling	18	1992	June 9, 2019
Bayesian inference in Stan Modeling	40	6326	July 18, 2017
Normal distribution with matrix mean and vector variance divergent transitions issues General rstan , techniques , fitting-issues	13	1001	February 11, 2021
Changing regression response variable from gaussian to logistic Modeling fitting-issues	30	1067	February 29, 2024
Observing non-linear function of latent variables Modeling	20	574	September 23, 2024

Variance model

Related topics