When running rstan, chain status messages print out every refresh iterations (I assume it's the same in CmdStan). Please add the option for elapsed time, number of divergent transitions, etc.
Description:
When running rstan, chain status messages print out every refresh iterations (I assume it's the same in CmdStan). It's hard to tell how long things might take, so having an option to print elapsed time would be very helpful. I know extrapolation of completion time is a very rough estimate, but it's better than nothing, and simply printing elapsed time puts the onus on us to do the extrapolation.
Adding the number of divergent transitions and, say, the acceptance rate or tree depth would allow us to terminate a run early, at our judgement. Personally, if I see a single divergent transition in Sampling, I'll stop it right there. Others may want to let it finish -- particularly if it will end in a short period of time -- then check some kind of diagnostics.
Summary:
When running rstan, chain status messages print out every
refresh
iterations (I assume it's the same in CmdStan). Please add the option for elapsed time, number of divergent transitions, etc.Description:
When running rstan, chain status messages print out every
refresh
iterations (I assume it's the same in CmdStan). It's hard to tell how long things might take, so having an option to print elapsed time would be very helpful. I know extrapolation of completion time is a very rough estimate, but it's better than nothing, and simply printing elapsed time puts the onus on us to do the extrapolation.Adding the number of divergent transitions and, say, the acceptance rate or tree depth would allow us to terminate a run early, at our judgement. Personally, if I see a single divergent transition in Sampling, I'll stop it right there. Others may want to let it finish -- particularly if it will end in a short period of time -- then check some kind of diagnostics.