need stepsize jitter?

[Bob Carpenter]

We have that in HMC now, but I think Michael wanted to get rid of
it. That'd be OK with me if nobody's using it.

- Bob

[Michael Betancourt]

It doesn’t really help — in cases where there is position-dependent
curvature the jitter would have to be huge to to get any kind of
implicit adaptation. And then it would be extremely wasteful as
most of the step sizes would not be optimal anywhere, and you’d
get stuck trying to randomly sample the few locally-optimal values.

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Bob Carpenter]

OK, my memory was finally correct! So let's get rid of
that "feature" going forward. I'd rather not have vestigial
features to doc and support and test.

- Bob