Kyurae Kim
Kyurae Kim
Hmm.. seems like a `mapreduce` diffrule is broken... again? @devmotion I think you're probably the expert on differentiating `mapreduce`. Do you have any clue about what's going on?
@torfjelde Hmm.. I don't think the hotfix fixes the MWE. In fact the MWE results in `N=2` so it doesn't hit the exit condition.
For anyone that hits this and needs to get this fixed ASAP, here's a quick and dirty solution (at the cost of twice the computation time...): ```julia @eval Bijectors begin...
Thank you so much for the PR. It looks pretty good at a glance, but I am currently on a trip to ICML, so it will take some time for...
Sorry for the delay! @arnauqb I don't have too much insight into the PPL part so that's something that perhaps @sunxd3 could comment on. @sunxd3 STL stands for sticking-the-landing, which...
Sorry for the delay! I just got back from ICML and was recovering from a cold I caught on the way back. Yes, it seems like we'll have to call...
@arnauqb The entropy under nonlinear transformations is generally intractable, so it's not implemented. For the `repgradelbo` case, for example, the jacobian adjustment part of the entropy is inside the Monte...
@arnauqb Can't you just shove `q_stop` into `reparam_with_entropy`?
Ah you're right. Okay this is indeed a little bit tricky, and I am not too happy with dispatching based on the objective. I think the best course of action...
Hi @arnauqb , did you push all your changes? If so, I'll take a deeper look so that we can move this PR forward.