Alex Lew

Good question -- it's really for support that's too large to tractably enumerate, not just infinite support. It is used whenever preferred values are used (it's the "other" token from...

Hi there -- thanks for your interest! To get started with Gen, we recommend following the 'Gen Quick Start.ipynb' notebook, which contains links to other notebooks in their intended order....

My 2 cents is I’d like to leave this (and multimixture) here, in the examples directory, as pedagogical examples :)

@fsaad 1) Should the implementation of importance resampling use log-categorical? 2) What should happen when importance resampling if all particles have zero weight? My instinct is that (1) should be...

OK, prototyped some support for this at https://github.com/probcomp/GenFluxOptimizers.jl. One question, though, is how to handle Gen's `ADAM` update configuration. It is not supported by the built-in modeling languages, only by...

Yep -- @SamWitty in case it wasn't clear from my messages, GenFluxOptimizers.jl is not a fork of Gen, but rather a package meant to be used alongside it. Just add...

I agree with all this, and will add that: 1. It could be useful to allow arbitrary generative functions to support constraining the return value, if they want to (e.g.,...

> Those two representations of a binomial sound like two different generative functions to me. To be clear, I totally agree --- I just mean that you might like to...

Yes, this would be great. I also think breaking the ratio down into the four components (forward and backward model and proposal scores) would be useful in a diagnostic-MH function.

@femtomc I don't totally follow — is the Kernel DSL still broken in some way? If not, can we rename this issue, and add some description of what problem it's...