I haven’t upgraded to PyMC v4, so that would likely be the source of the error.
With regard to the penalization, I read the part you mentioned in that paper where the authors refer to an IID normal prior being equivalent to a Frobenius norm on A. I didn’t know that ahead of time but it’s not surprising to me that it’s the case - it looks like you can omit the Frobenius penalty from the code I wrote.