Dear brms and cmdstanr users,
A brms program is running on my Mac now. I use cmdstanr through backend of brms. The sampling is running properly but percentage of the progress is presented in very small steps like:
Chain 4 Iteration: 20800 / 50000 [ 41%] (Sampling)
Chain 4 Iteration: 20900 / 50000 [ 41%] (Sampling)
…
Can I change the scale of the progress percentage presentation? I’m happy if it is 10%-step.
This is so trivial but so many lines in console is inconvenient…
Thank you very much.
environment
macOS Catalina version 10.15.7
R 4.1.1
cmdstanr version 0.4.0
brms version 2.16.1.
Rstudio Version 1.4.1106
Use for example refresh=5000
to print progress every 5000 iterations.
2 Likes
For what it’s worth, it’s very rare that you would need 50000 iterations from Stan.
2 Likes
I missed it in the manual… Thank you very much!
I’m a beginner of Bayesian statistics and some texts I found said relatively large size of iterations is needed to calculate Bayes factor. Probably 10000-iterations is enough but some studies employed 20000 iterations. So I just think 50000-iteration is surely enough for any case. But as you commented, it seems weird-level of iterations. I should reduce it to around 15000. Thank you very much for your comment!
The Bayesian text that you have read may have used simpler MCMC algorithms like Metropolis-Hastings. Then the samples have a lot of autocorrelation and it is understandable that a lot of iterations are needed. With Stan the effective sample size is higher so less iterations is needed, I think this is what @jsocolar meant.
I did not care about algorithms in texts and previous studies. Seems I need check it to understand properly the number of iterations. Thank you very much for your comment. I really appreciate it.