Saturating the max tree depth for every sample in extremely simple model

[syclik]

@jwolcott, no prob!

@jwolcott

Yeah, I understand. I was kind of hoping that somebody would just recognize the symptoms and be able to give me a conceptual answer that points me in the right direction. But I do appreciate your effort nonetheless.

Yeah… that’s usually the right way to go about it. Unfortunately, this problem is in the details. Just based on what output you’ve shown, it looks like the autodiff isn’t actually working. Maybe trace through and make sure everything is using the correct types?

@jwolcott

Unfortunately the code isn’t public. I’m not trying to be cagey, but it’s part of a large framework that I don’t have the rights to ‘declassify’ and even if I could my guess is that I’d be wasting your time digging through a class hierarchy that’s largely not relevant to the problem.

Understood.

@jwolcott

I’m going to try to write up a simpler test case that exhibits the same behavior which I can share here so that it’s straightforward to look at. Hopefully that won’t take me more than a day or two.

Great. If you look in Stan’s unit tests, there are simple tests that demonstrate calling the different functions. Maybe that’ll clue you in on what’s happening?