Include "transformed data" to argument of az.from_pystan()

[HJA24]

Hi,

One of the arguments for the function az.from_pystan() is "observed_data". According to the docs "observed data" is extracted from the posterior.data.
My Stan model uses the "transformed data"-section extensively and I would like to extract variables defined here intoaz.InferenceData-object.

Is it possible to make sure that also the data from "transformed data" can be selected for "observed data"?

The corresponding error-log:

File "/Users/HJA/PycharmProjects/ufc/scripts/sample.py", line 58, in sample
    return az.from_pystan(
  File "/Users/HJA/opt/anaconda3/envs/ufc/lib/python3.10/site-packages/arviz/data/io_pystan.py", line 1079, in from_pystan
    ).to_inference_data()
  File "/Users/HJA/opt/anaconda3/envs/ufc/lib/python3.10/site-packages/arviz/data/io_pystan.py", line 609, in to_inference_data
    obs_const_dict = self.observed_and_constant_data_to_xarray()
  File "/Users/HJA/opt/anaconda3/envs/ufc/lib/python3.10/site-packages/arviz/data/base.py", line 65, in wrapped
    return func(cls)
  File "/Users/HJA/opt/anaconda3/envs/ufc/lib/python3.10/site-packages/arviz/data/base.py", line 65, in wrapped
    return func(cls)
  File "/Users/HJA/opt/anaconda3/envs/ufc/lib/python3.10/site-packages/arviz/data/io_pystan.py", line 573, in observed_and_constant_data_to_xarray
    vals = np.atleast_1d(posterior_model.data[key])
KeyError: 'y1'

[aloctavodia]

@ahartikainen can you check this?

[ahartikainen]

Are we talking about pystan 2.x or 3.x? I need to refresh my memory (I nowdays mostly use cmdstanpy), but if the transformed data is available then we can pull it from the sampling object.

[HJA24]

@ahartikainen Pystan 3.x