Seth Axen

> Should this worry us/potential users? I don't think so. That being said, computing the maximum energy error over the whole macro-step would be pretty cheap (with each micro-step, track...

Integration tests fail due to AdvancedHMC v0.7.0 replacing `rand` with an internal function `rand_momentum`. I've opened TuringLang/AdvancedHMC.jl#411 to discuss making `rand_momentum` part of the API so we can safely overload...

In my view it would make more sense to implement such a handler as a new struct implementing the LogDensityProblems interface in its own package. Then one could track this...

Depends on how you want to implement it! You could e.g. just implement an object that has a callback, which would then support both modifying how the original object works...

Ah, cool. Yeah actually if you do implement something that would allow for custom handling of errors, etc in its own package. I'd actually drop the `nan_on_error` field here and...

The basic functionality is here. All that remains is to remove now-unused methods and their corresponding tests.