model.stan (2.2 KB) Hi there! :wave: I’m new to stan and I could use a bit of feedback on this model. Is there anything big I’m overlooking? Does something stand out to you? data { int<lower=1> J; // number of apps int<lower=3> L; // number of years int<lower=1> N; // number of observations…

[image] statwonk: Divergences can be caused by non-sensical priors Divergencies can occur when you need a hyperparameter but instead have hard-coded a parameter Those are both instances of model misspecification. [image] statwonk: One of the most important things to do is “center” or “s…

Thank you Bob. These are all very helpful clues and give me new avenues to learn / improve the model.

[image] statwonk: I’m new to stan Welcome on board! Great start with your model. I agree with Bob’s tips. On the question of scaling or standardizing, I prefer the former since scaling by a constant doesn’t assume the mean and sd are known a priori. In rstanarm the default is to do some res…

[image] jonah: It’s basically an approximation to making the model hierarchical and learning the values from the data instead of assuming them. Neat—hadn’t thought of it that way before.

In case it’s of interest, here’s a thread that has some other views on related topics: [image] Standardizing predictors and outputs in a hierarchical model Modeling Hello everyone, First of all thanks for the StanCon—it was a great experience. Here is a question that I …

Thank you both! I’d say so far the parameter centering and choosing priors was the hardest part for me to figure out. This final comment is really interesting: Standardizing predictors and outputs in a hierarchical model Okay so my understanding is that my original model with the non-centered para…

So strange I’m hitting this error when I up the dataset using the transform parameters centering trick (it’s still fairly smallish N ~ 15k), On Ubuntu 14.04, I get Error in unserialize(socklist[[n]]) : error reading from connection — sounds related to: Unserialize error I wasn’t running into the s…

I was able to verify that that the serialization error above comes is due to being out of memory. I used the free command to log memory usage and sure enough I ran out, nohup watch -n 1 'free -mh' > mem.log & I noticed in another thread that either Bob or Ben said rstan allocates memory up front w…

You can also stream out in R. What I’ve never understood is whether you can turn off saving in memory. You can also do fewer draws (rarely does anyone need more than a couple hundred effective sample size) or save fewer parameters.

I don’t think that’s true. I tried the sample_file arg on stan but memory still ballooned. Since standardizing in transform parameters adds a significant amount of overhead and most of my priors are centered between 0 - 5 with standard deviations of +/- 2 does standardizing buy me much? Could I mo…

Looking for a stan program critique

Modeling

jonah April 14, 2018, 1:15am 6

In case it’s of interest, here’s a thread that has some other views on related topics:

Topic		Replies	Views
NLME model in STAN for missing + unbalanced longitudinal data Modeling rstan , mixed-model , missing-data	20	651	October 21, 2024
Non-Centered Parameterization - Divergence issues w PyStan hierarchical model Modeling specification , hierarchical-model , divergences , reparametrization	3	852	April 4, 2022
Multivariate Hierarchical Model Modeling techniques , fitting-issues	7	1076	April 7, 2021
Cross-classified Hierarchical Model, Heterogenous Variances Modeling	19	1712	August 11, 2018
Posterior of my non-centered parameterization (of intercepts hierarchical model) is less navigable than my centered parameterization Modeling rstan , fitting-issues , specification , biology	5	547	February 24, 2024

Looking for a stan program critique

Related topics