Behaviour of `transformed_data` in `from_cmdstanpy` converter

[OriolAbril]

Stan programs have a set of defined blocks like data, parameters or model. from_cmdstanpy has two observed_data and constant_data that take dictionaries of string->array-like, and data in a similar fashion is a required input when sampling with cmdstanpy. But there is also a transformed data block. Users can obviously transform the data manually and add it to the respective dict. But if there is a way to have cmdstanpy handle the transformations I think it would be nice to avoid this duplication.

An option that comes to mind could be adding {"key": None} to signal the value should be extracted/generated from the transformed data block.

How do other people feel about this? Anyone knows if it is possible? Other better ideas for the API?

[patelshivani2283-lab]

Hi @OriolAbril,

I would be interested in exploring this issue.

Before thinking about API design, I wanted to clarify:

• Does cmdstanpy expose the transformed data block separately in the fit object?
• If so, would it be possible to access it similarly to observed_data?
• Or would this require changes on the cmdstanpy side?

I can look into how cmdstanpy stores transformed data internally and report back.

Thanks!

[ahartikainen]

I don't think cmdstan / cmdstanpy support this?