_rng in transformed data block

[Krzysztof Sakrejda]

I've been writing some scripts for ODSC Boston that simulate data using model parameters or the generated quantities block. The manual says the _rng functions can only be used in generated quantities but it would be nice to simulate some discrete values and then use the model block to simulate from a relevant posterior. I guess for simple models I can pack everything into the generated quantities section but that won't work if I want to simulate from something more complex (e.g.-inhomogeneous poisson process with a smooth intensity function).

The question here is whether this is straightforward but we don't have time for it or whether there's a deeper reason we're not doing it. Sorry for the barrage of simple questions but I've been exercising Stan usage more than usual for this workshop.

Krzysztof

[Ben Goodrich]

I believe that we agreed this was a good idea and that it was fairly easy to do and haven't done it.

 

[Krzysztof Sakrejda]

On Thursday, May 12, 2016 at 9:43:13 PM UTC-4, Ben Goodrich wrote:
[snip]

> I believe that we agreed this was a good idea and that it was fairly easy to do and haven't done it.
>
>  

Thanks for looping me in. Seems like a great idea but it's not on my short list either! Krzysztof

[Bob Carpenter]

Could somebody remind me what the use case is?
I wasn't sure everyone thought it's a good idea.

One issue is not being able to save the randomization.
Or maybe we add an argument to the interfaces to dump
the transformed parameters?

- 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.

[Sebastian Weber]

+1 for dumping the transformed parameters optionally. The reason is that I write often many stan functions which rely on data processing in the transformed data block. Currently I have to redo those transformed data steps in R such that I have the right input to the Stan functions.

Sebastian

[Bob Carpenter]

That's not at all related to _rng in transformed data. Is
this the right thread? We'll probably be moving to saving only
a subset of the parameters in Stan 3, but we can't seem to
find the time to ever get there.

- Bob

> On May 13, 2016, at 12:12 PM, Sebastian Weber <sdw....@gmail.com> wrote:
>
> +1 for dumping the transformed parameters optionally. The reason is that I write often many stan functions which rely on data processing in the transformed data block. Currently I have to redo those transformed data steps in R such that I have the right input to the Stan functions.
>
> Sebastian
>

[Krzysztof Sakrejda]

On Friday, May 13, 2016 at 3:04:11 PM UTC-4, Bob Carpenter wrote:
> That's not at all related to _rng in transformed data. Is
> this the right thread? We'll probably be moving to saving only
> a subset of the parameters in Stan 3, but we can't seem to
> find the time to ever get there.

Don't loose hope, it'll happen! K