Rob Zinkov

> Would something that can also update existing coords be more useful? Perhaps named `set_coords`? Maybe, but updating coords likely would require something like the tricky shape-checking logic in `set_data`....

I don't think `sample_posterior_predictive` needs to be made more restrictive. That the posterior predictive is usually used for replications of the observation data is more of a convention. There is...

It doesn't need to be passed in, and I'm equally happy with an explicit input variable. Especially if the error messages when forgetting to include a trace are helpful. But...

I think articulating what model variables are ignored is also something that can be better documented. Since my mental model right now is, all variables that were not observations present...

> I remember that I didn't like this pattern before, because it messes up the typing of the arguments =/ > > I know that the typing of distribution arguments...

> This is a fun and interesting idea, and I really enjoy the inventiveness of the implementation! > > However, I'm not sure I really would want this to be...

> Storing and loading results with unnamed TensorVariables as the keys would be a mess. I think there is a small mess that comes from needing to hold onto the...

> Will this work for unicode variable names? People have a habit of using unicode mu, sigma, beta, etc. If yes, it's worth adding some tests for that. > >...

> I don't think there is any benefit to the current approach, it's just the best we could do within Python's limitations (other than going the JAGS way, and defining...

That should be workable.