I would like to extract the warmed up / sampling parameters using cmdstanr, in particular the step size. (I already have the inverse metric using fit$inv_metric()). After browsing Function reference • cmdstanr , I couldn’t find any function to get the sampler parameters. I know how to do this with r…

Ah, here it is fit$metadata() which included the adapted step size!

Indeed, fit$metadata() includes everything that is inside the result CSV (even if you modify cmdstan and print some additional info in the CSV by yourself), apart from inv_metric which has its own method.

Will check out and I’ll compare restarting the chain and just running the full chain (with rstan, I didn’t see any difference between the two). Related question: when using fit$sample(), are the samples ordered? I know they’re not with rstan, unless I include the warmup.

They appear in the same order as in the CSV output.

I found it was sensitive to the model and data, try it with the data/model generated by the gist near the end .

For posterity, the bug was worked out, and I also posted some code that achieves things properly.

Extracting step size per chain in cmdstanr

Interfaces

mike-lawrence January 12, 2021, 2:26pm 4

Make sure to see here for ongoing exploration of a bug associated with using those values to re-start a chain:

Topic		Replies	Views
Sampling chains from the middle in cmdstanpy General	10	567	October 8, 2020
Passing a different mass matrix for each chain in cmdstanr Interfaces	7	771	January 13, 2021
Resume sampling after interuption Interfaces cmdstanpy	5	1409	May 12, 2020
Continue Warmup / Sampling after Interruption General	10	2840	July 31, 2022
Saved mass matrix General	22	2906	December 12, 2022

Extracting step size per chain in cmdstanr

Related topics