Adrien Corenflos
Adrien Corenflos
Delayed rejection is a fairly old idea, there may be some room to make it a meta-algorithm rather than an instance of HMC. See https://www.jstor.org/stable/2673700
The HMM sampling can't really be done using prefix-sum. You would need to use a divide and conquer approach as in Section 3.2 in my paper https://arxiv.org/abs/2303.00301 (Note that it's...
The LGSSM is unsurprising (it's just a cumulative sum over pre-generated Gaussians). In fact I think I had it implemented some time ago. I'm actually surprised by the HMM case...
I don't see how you would do data tempering efficiently within JAX but apart from this I'm okay with the suggestion.
I'd be happy to review a PR on this. I think it's a valid contribution too. I'm not an expert on nested sampling but also not clueless so hopefully it...
Apologies, I was travelling, I will check this ASAP
c.f. my answer, I'd like for use to confirm there is a problem/where the problem is before designing a fix.
I'm not sure what Rémi meant by this issue. JVP vs VJP seems like a JAX-only problem to explain and is a bit orthogonal to BlackJAX. Additionally, most of the...
> I want this because I have to differentiate through an ODE solver, and this is faster with forward-mode automatic differentiation for my use case. This is just very surprising...
Same here, I had planned to do it at some point but I've not been able to commit the time to :D very happy someone is doing it! Design choice...