Sampler Diagnostics

[Raoul-Kima]

@bbbales2

As far as I know the only thing Stan does with a divergent transition is mark that it is divergent. It still samples from the trajectory in the same way as it always would.

Betanalpha wrote:

When a divergence is identified the dynamic trajectory expansion terminates and a new state is chosen from the trajectory that has been build up to that point

So the divergence does influence how the sample is chosen.

The quote is from here:

Getting the location + gradients of divergences (not of iteration starting points)

When a divergence is identified the dynamic trajectory expansion terminates and a new state is chosen from the trajectory that has been build up to that point – it does not just take the initial point. But more importantly as @sakrejda notes, a divergence is not a point but a property of the trajectory itself. There is no discrete threshold below which the trajectory is fine and above which it’s bad – it’s the entire trajectory (really the entire modified energy level set) that’s pathological.…

I’ve now solved this problem, at least provisionally, by changing your oneSampleNuts function to check for divergence in addition to U-turn before merging in a new subtree. I hope this is close to how Stan does it.