set_data raises AttributeError

[ghost]

Description of your problem

After creating a model that uses pm.Data objects for all data inputs, using set_data following sampling to swap in new predictor values for prediction results in an AttributeError

coords = {
    'age': ages.values.astype(int),
    'obs_id': np.arange(df_clean.shape[0])
}
with pm.Model(coords=coords) as age_model:

    age_ind = pm.Data('age_ind', age_indices, dims='obs_id')

    ...


with age_model:
    pm.set_data({"age_ind": nola.age.values - 20})
    post_pred = pm.sample_posterior_predictive(trace)

    ...

Please provide the full traceback.

Complete error traceback

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-67-459704977541> in <module>()
      1 with age_model:
----> 2     pm.set_data({"age_ind": nola.age.values - 20})
      3     post_pred = pm.sample_posterior_predictive(trace)

1 frames
/usr/local/lib/python3.7/dist-packages/pymc/model.py in set_data(new_data, model)
   1759 
   1760     for variable_name, new_value in new_data.items():
-> 1761         model.set_data(variable_name, new_value)
   1762 
   1763 

/usr/local/lib/python3.7/dist-packages/pymc/model.py in set_data(self, name, values, coords)
   1175             # NOTE: If there are multiple pm.MutableData containers sharing this dim, but the user only
   1176             #       changes the values for one of them, they will run into shape problems nonetheless.
-> 1177             length_belongs_to = length_tensor.owner.inputs[0].owner.inputs[0]
   1178             if not isinstance(length_belongs_to, SharedVariable) and length_changed:
   1179                 raise ShapeError(

AttributeError: 'NoneType' object has no attribute 'inputs'

Versions and main components

• PyMC/PyMC3 Version: 4.0.0b2
• Aesara/Theano Version: 2.3.4
• Python Version: 3.7.12
• Operating system: Linux (Colab)
• How did you install PyMC/PyMC3: pip

[ghost]

This appears to be due to my using a dims argument on the original Data which does not correspond to the new data. At any rate, there should be a more informative error message here.

[ricardoV94]

CC @michaelosthege

[michaelosthege]

I remember that we changed something related to this, but it looks like we didn't link this issue.

import numpy as np
import pymc as pm

print(pm.__version__)

coords = {
    'age': np.arange(3),
    'obs_id': np.arange(5),
}
with pm.Model(coords=coords) as age_model:
    age_ind = pm.MutableData('age_ind', [1,2,3,4,5], dims='obs_id')


with age_model:
    pm.set_data({"age_ind": [6,7,8,9,10]})
print("green")
# 4.1.7
# green