discuss Langevin

[paciorek]

I had done some work a while back to implement a version of Langevin (in nimble branch extend_langevin) that adapts the full covariance following a discussion with Giacomo Zanella (https://arxiv.org/abs/1908.11812). I hadn't gotten acceptable performance with that or (I believe) with the basic Langevin on my test cases and hadn't gotten a chance to get back to it to explore further.

I'm wary of exposing the current Langevin in nimbleHMC until we do more exploration of performance/behavior. One thought is that we remove it entirely for now, while another is that we don't export it.

@danielturek @perrydv, thoughts?

[danielturek]

@paciorek I'd be fine with the non-export option.

Also noting that the nimbleHMC repo README says:

• Langevin sampler (under development)

[paciorek]

Great, I'll leave it to you to do that.

[danielturek]

sampler_langevin is no longer exported in the nimbleHMC package.

I'm leaving this issue open, for discussion about re-working / improving the langevin sampler.

[paciorek]

Revisiting this in light of upcoming release.

sampler_langevin is not exported but there is roxygen and one does see it if one does library(help=nimbleHMC). I'm wary of this. I think my vote at this point would be to remove the roxygen for it and make sure it doesn't show up in man.

@danielturek how does that sound?

[danielturek]

@paciorek Yes, that sounds reasonable to me. I'm happy to make this change (remove Roxygen for sampler_langevin) if we're in agreement here.

[paciorek]

'removal' could also simply be "commenting out" the roxygen "comments" by adding another level of comment hashes, so as to keep the info in the code file.

[danielturek]

@paciorek Yes absolutely, that was my intention. I'll do this right now. Keep this issue open, though.