Error in Forecasting_with_structural_timeseries (too many indices for array)

Forecasting with Structural AR Timeseries

Issue description

Executing step by step each cell I got: too many indices for array at line:

IndexError                                Traceback (most recent call last)
<ipython-input-13-fc80453dcf5f> in <module>
     15         sigma=sigma,
     16         constant=True,
---> 17         dims="obs_id_fut_1",
     18     )
     19     yhat_fut = pm.Normal("yhat_fut", mu=ar1_fut[1:], sigma=sigma, dims="obs_id_fut")

Note: I'm not sure if the bug is related to pymc it self or the example, I can't find a way to verify that.

Expected output (according to the doc)

Sampling: [ar1_fut, likelihood, yhat_fut]

 100.00% [8000/8000 00:31<00:00]

Additional info:

Execution environment:

• python v3.7
• Miniconda + custom pip packages
• BeakerX (jupyter notebook).

Proposed solution

Probably reducing dimension.

Open the linked Forecasting with Structural AR Timeseries notebook and reproduce the failing cell at the dims="obs_id_fut_1" line. Compare the notebook's behavior with the reported Python 3.7 and BeakerX environment to determine whether the problem is in the example or PyMC. Done means the documented sampling output is produced or the compatibility issue is clearly identified.