Colin Carroll

Hey @jeffpollock9 -- I used some of the ideas from your `_transform_reshape_split` bijector in the [new windowed sampler](https://github.com/tensorflow/probability/blob/master/tensorflow_probability/python/experimental/mcmc/windowed_sampling.py#L93) -- (re) using some of that code might be helpful here.

Specifically, this is coming from using a `JointDistributionCoroutine`, which dynamically generates a StructTuple to represent the state, which means `pickle` can't represent it. Depending on what kind of object `models['states']['birth_rate']`...

Hey! I took a quick look at this, and a few things: 1. running the code snippets you provided does a _better_ job of fitting the parameters than the image...

Ah, right, forward sampling will give you properly signed parameters, but notice that you are directly evaluating the probability of `intensity`, which must be positive. This turned up for me...

That's pretty interesting! The math works _formally_ with any real number, I think, and also the stochastic root finding will do its best to find the value of step size...

I'm also a little distant from the code here, but from that short distance, it would be neat to have ghmc and an ECA framework, so that implementing MEADS was...

Not super helpful, but @aseyboldt helped me get `nutpie` working with JAX in `bayeux`: https://github.com/jax-ml/bayeux/blob/main/bayeux/_src/mcmc/nutpie.py Broadly, `bayeux` accepts a log density in JAX, then uses function transforms to compute gradients,...

Are you running at HEAD or with the latest pypi release? I _think_ the first should work (https://github.com/jax-ml/oryx/pull/98), but the second shouldn't. I can make a release sometime soon if...

I'll put in a vote against this -- there are enough pymc scripts and libraries out there that _don't_ follow library updates very closely, but would like to continue to...

Also https://github.com/jax-ml/jax/discussions/24501 On Mon, Oct 14, 2024, 8:23 AM Junpeng Lao ***@***.***> wrote: > related: pyro-ppl/numpyro#1867 > > likely rootcause and workaround see: jax-ml/jax#23822 > > > — > Reply...