Can't load MMM with vector parameter prior

[ricardoV94]

Grabbing the current example in #324 , and changing the beta prior to:

custom_beta_channel_prior = {'beta_channel': {'sigma': [1, 2], 'dims': ('channel',)}}
my_model_config = dummy_model.default_model_config | custom_beta_channel_prior

I then fail to load the model after saving with:

ValueError                                Traceback (most recent call last)
Cell In[63], line 1
----> 1 loaded_model = DelayedSaturatedMMM.load('my_saved_model')

File ~/Documents/Work/Labs/pymmmc/pymc_marketing/mmm/delayed_saturated_mmm.py:416, in BaseDelayedSaturatedMMM.load(cls, fname)
    414 # All previously used data is in idata.
    415 if model.id != idata.attrs["id"]:
--> 416     raise ValueError(
    417         f"The file '{fname}' does not contain an inference data of the same model or configuration as '{cls._model_type}'"
    418     )
    420 return model

ValueError: The file 'my_saved_model' does not contain an inference data of the same model or configuration as 'DelayedSaturatedMMM'

The error doesn't happen if I use a scalar sigma.

CC @michaelraczycki

[michaelraczycki]

It's something that comes from json encoding. It's impossible to determine if json list was encoded from np.array, or from list. Because of that load automatically creates np.array in place of each json list found in the model_config. I see that I need to create full documentation of the model builder module and it's retirements/limitations.
Hashlib.sha256 encodes differently python list and numpy array, and because of that Id of the model doesn't match after loading. In my opinion we should just require passing that on form of np.array format always (then it won't cause the error

[ricardoV94]

What about sanitizing the config (after making a copy) so the user doesn't have to worry about that? In this case convert lists to numpy arrays.

[michaelraczycki]

that can also be done, although I think I'd suggest to first release the intro notebook, and then start patching up those topics as the jsonification looks like a bottomless pit, probably every once in a while we'll get a bug report about that.

[michaelraczycki]

I'll open another issue in pymc-experimental to keep track of that

[michaelraczycki]

I've opened a PR that now will convert any lists or arrays found inside the model_config into np.array format before assigning model_config inside the DelayedSaturatedMMM, probably merging #351 should solve your problem