PRNGs in transformed data

[Bob Carpenter]

This is worth its own thread.

The only reason we're not doing this now (that I recall) is that
there's no way to save the results of the transformed data
other than as a generated quantity, which is super inefficient. Or
by keeping track of the seed so you can replicate. If that's OK, it'd
be trivial to allow PRNGs in transformed data. As in it would take an
afternoon, including testing and updating the doc. It'd be much
more challenging to dump out the results and we'd want to wait and
make this part of Stan 3, I'd think. Or we could allow it now with
no way to save and add saving the output to our to-do list.

- Bob

[Ben Goodrich]

Can we expose a print_csv() function or something that would allow the user to copy and paste from the stream formerly known as stdout?

I think it would be worth doing soon even without a great way to save results just for the ability to test that you can recover parameters that you pass in as data.

Ben

[Bob Carpenter]

OK, I can add PRNGs in transformed data if there aren't any
persuasive objections. I created an issue:

https://github.com/stan-dev/stan/issues/1644

As to print_csv(), are there arguments and what's printed? I
didn't understand the motivation, either --- is it to dump out
the randomly generated transformed parameters somehow? We
have access to an output stream that we need for print(), so
the plumbing's all there. In fact, you could probably write
a print_csv() as a user-defined function.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.