Checkpointing CmdStan sampling

Thank you for providing clarification on this, wds15.

I’m bumping this because we aren’t able to complete warmup within a walltime limit, but I can’t find where in the stan code the 3-window warmup schedule is implemented. I mean, I can find the cpp files for the windows sizes, step size adaptation but nothing about how the windows could be translated to cmdstan arguments. Is this documented somewhere?

See

And user guide in assets https://github.com/stan-dev/cmdstan/releases

Let’s see if we get everything we need from these urls

1 Like

Thanks, that’s what I was looking for. From that I start to see how to write a script around a CmdStan model which schedules that by hand, which would let us sidestep our walltime limitations.

@maedoc Did you ever get this sorted out? I’m looking at a similar problem and trying to decide if it’s worth the effort.