Luciano Paz

@alexlyttle, we're thrilled that your model achieves such a big speedup with pymc4 (and GPU). Could I ask you a few questions? How big is your model, with this I...

Thanks for the feedback @alexlyttle. About publishing, you have to be aware that at the moment, pymc4 has no initial tuning for the NUTS sampler. This means that the mass...

@OriolAbril, ok so we will have to write custom `save_trace` and `load_trace` functions in pymc4 that change the `/` character under the hood to something that doesn't break net_cdf.

We plan to allow a single parametrization in the each distribution instance's initialization function. Pymc3 supported multiple parametrizations in `__init__` (e.g. the `Normal`) and that made things harder to maintain....

@Sayam753, you're right. Due to our confusing usage of the operation, we should have named it `LogOdds` instead! We can open that in another issue.

@ferrine, what would be an updated status on the tasks here? Some refer to the multi backend things that were discarded later. The executors now work and are tested but...

The reason that we cast to int32 and not to int64 is that the latter can overflow float64 calculations for large integer values, whereas int32 won't. You could try to...

Thanks for opening the issue @cbrummitt. I had completely lost your question over on discourse. In [v4 we wrote a new way to do posterior predictive sampling](https://github.com/pymc-devs/pymc/pull/5759) that is flexible...

Hi everyone, first of all, thanks for your amazing work! I came across this issue today because I have a dataset with multiple variables and multiple multi index dimensions, some...

Yeah, we hadn’t thought about the sharedvalue coordinates when the designed the volatility propagation rule in sample posterior predictive. We will have to change it slightly. If the user supplies...