Stan warm-up/initialization phase

Interfaces CmdStan
1

Hi there, I am very new to Stan and wanted to get insight from all experts in this great community.

How should I find the time spent in warm-up period/ Stan initialization phase? In my model, when I set a weaker prior for my parameter, Stan run time increases exponentially !! It is interesting to me to see such effect and I wanted to see how initialization time changes when I change my prior to a stronger/more-informative one. And what is a workaround for that?

Should I check ‘stepsize__’ and ‘n_leapfrog__’?

PS: I am using cmdStan

2

The time spent in each phase should part of the stansummary output.

The fact that priors change your running times a lot is an indication that the weaker prior allows the sampler to move parts of parameter space that have complicated geometry - most likely, the model is not fully identifiable and the prior helped it. It is definitely a reason to be concerned whether the model is specified correctly, but hard to say anything without seeing the model.