Horseshoe prior in rstan

[avehtari]

@Se_Yoon_Lee

I check stationarity by see the traceplot in my naked eyes.

That is not sufficient. If you are using RStan you can use monitor function.

@Se_Yoon_Lee

For simulation study, I typically use n = 100 p =500, but in gene expression data study, obviously p is much larger than 500, saying 5000.

See here for an example with p>5000

What's the highest dimensional model Stan can fit using NUTS?

Program and data are not secret. I was just in a hurry when writing that post. Data is Prostate_GE listed at http://featureselection.asu.edu/datasets.php The model is described in Piironen and Vehtari (2017) https://projecteuclid.org/euclid.ejs/1513306866. The code is slightly modified from appendix C.1. There are less transformed parameters saved and bernoulli_logit_glm() is used instead of bernoulli_logit() (this gives about 4x speedup). bernoulli_logit_glm_rhs.stan code is data { int<lo…

and the comments about the speed What's the highest dimensional model Stan can fit using NUTS? - #5 by avehtari
With my 3+ year old laptop it’s 2 hours, with GPUs it would be less than 40mins