Adrian Seyboldt comments

Results 186 comments of


                                            Adrian Seyboldt

Allow multiple datasets in fit_to_data and add option to return opt_state

How about this version? This avoids the breaking change by having the old `x` argument and the extra positional arguments. I'm not sure what to do about the `condition`/`weights` problem....

Allow multiple datasets in fit_to_data and add option to return opt_state

Didn't mean to close...

Allow multiple datasets in fit_to_data and add option to return opt_state

I like your version :-)

nutpie worker panic under jax backend

Sounds like pytensor might be doing something strange here? Maybe somewhat related to this? https://github.com/pymc-devs/pytensor/issues/526 Does this also happen with `gradient_backend="jax"`?

nutpie worker panic under jax backend

I've seen several examples where the pytensor gradient was significantly faster. Jax also sometimes has extra nan issues, that pytensor avoids with rewrites ( https://docs.jax.dev/en/latest/faq.html#gradients-contain-nan-where-using-where ) If you can figure...

Add inverse_and_gradient method to bijection

Thank you for having a look. I was seeing a much larger improvement in my earlier benchmarks, but turns out I made a mistake in those, and funnily enough there...

`local_subtensor_merge` can complicate graphs

A problem is that I think some scan rewrites depend on this. But I think we should probably restrict this one to static indexes or so and worry about fixing...

Several keyword arguments in `nutpie.sample` appear to do nothing

Did you figure out what the problem with store_unconstrained was?

Several keyword arguments in `nutpie.sample` appear to do nothing

You might want to have a look at `trace.sample_stats.unconstrained_position` ;-)

Several keyword arguments in `nutpie.sample` appear to do nothing

If `store_unconstrained=True`, it will store the parameter vector as one array in the sample stats. It doesn't change anything about `trace.posterior`, that will still contain only the constrained values. Maybe...