Michael Osthege

Every solution other than returning a new model will have nasty side effects on the current model. But this may actually result in an okay API: ```python with pmodel.set_data({ "x"...

@larryshamalama @ricardoV94 how much work is left on this one? Can we include it in the 4.0.2 milestone?

@bwengals @lucianopaz ths looks like either the refactoring of `MarginalApprox` to use the new `DensityDist` was incomplete? My hunch would be that both issues described above weould be resolved by...

I'm not convinced. What valuable information exactly is lost? Can you give an example? Also, with the current API (tuples) one can do `pmodel.coords["city"].index("London")` which is incredibly useful when building...

Sorry for the delayed response, I was pretty busy this week... No doubt that `pd.Index` is really useful for people who know the tricks, but I'm worried that it could...

> If you think live preview of the trace while it is sampling is important, I think we can find a solution for that, that is much less involved. This...

Oh and just to be clear, I think our intentions are well aligned: * Making working with `dims`, `coords` more capable & robust * Promoting good Bayesian workflow practices (with...

The whole implementation for switching the data in an existing model is broken. Shape issues are one thing, but @ricardoV94 is right that with imputation it could get even worse....

I have a hunch that we'll have to revisit the whole imputation feature under the new `RandomVariable` paradigm. After merging #4439 I'm fine with doing `observed=pm.Data(...).container.data` as a workaround. Let's...

> > > Agreed 👌 > Out of curiosity, what do you mean by "doing observed=pm.Data(...).container.data as a workaround" ? This way I can use the `pm.Data` to include my...