Suppose parameter \alpha is estimated using Model 1. If we have only one chain with a length of 2000 samples, then the posterior sample of \alpha has the dimension 2000 * 1 in stan output. I want to use this parameter as a constant value in Model 2 (number of chains and samples like Model 1). I do n…

In principle you can pass the samples from one model as data to the next model, whether or not this is reasonable depends on the question you are trying to answer. Personally, I would not know when this makes sense, but my creativity is quite limited. Maybe someone else knows?

[image] Funko_Unko: In principle you can pass the samples from one model as data to the next model, How can do this? can you give me an example?

In principle you can do this, but you probably should not: Model 1: parameters { real x; } model { x ~ normal(0,1); } Model 2: data { real x; } parameters { real y; } model { y ~ normal(x,1); } And then pass the samples from model 1 one at a time to model 2. However, I suspect you might a…

Thank you. But my problem is that x in the output of the first model is a vector as long as the number of samples and not a real number. Your latest model is fine, but I refer you to the question I asked on another topic .

This topic might be helpful to you: [image] Use posterior predictive distribution in another model Modeling Hello, I have individual-level predictions produced by a non-parametric Bayesian model. Specifically I have, for each observation, a posterior distribution of the in…

This thread might also be of interest.

For example, model_1 <- " data { int A; vector[A] x; } parameters { real mu; real<lower = 0> sigma2; } model { mu ~ normal(0,10); sigma2 ~ inv_gamma(2.1,2); x ~ normal(mu,sqrt(sigma2)); } " I want to use mu in below: model_2 <- " data { int A; vector[A] y; ???? mu; } parameters { real la…

@Funko_Unko ’s first option is to fit your model two iteratively, using a different posterior draw of mu as input data each time. Is your question about how to write the appropriate for-loop to do this? Note that it involves fitting model 2 many times, and so it might be slow. @Funko_Unko ’s second o…

first option: You are right, I do not want to fit model 2 many times. second option: How can I write my model with this strategy? Because in the @Funko_Unko ’s comment, x and y are parameters, while in my model, they must be data and mu and lambda are parameters. third option: Can you give a sim…

Second option: This involves fitting model 1 and model two jointly at the same time. Model 1 is re-fit here, so mu is a parameter. @Funko_Unko ’s x and y are just dummy names for model parameters. They could represent your mu and lambda. Third option: Suppose that you conclude that the posterior…

Use parameter output in the first model as an input in the next stan model

General

jsocolar May 11, 2021, 8:23pm 6

This topic might be helpful to you:

Topic		Replies	Views
How to pass parameter estimates from one model to another at every MCMC iteration? Modeling	12	1903	February 6, 2018
Use derived quantity from one model as response variable in another Modeling rstan , techniques	8	595	July 29, 2022
Use posterior from one model as data for another? Modeling specification , brms	5	1668	September 12, 2022
Models on models within stan Modeling rstan , specification	1	496	September 23, 2021
Using posterior samples as data for next model Modeling techniques , fitting-issues , state-space	5	1229	May 27, 2020

Use parameter output in the first model as an input in the next stan model

Related topics