I’m running Theano 1.1.2 and PyMC v4.0.0b2
I can try downgrading PyMC and see if that solves the problem.
Another question I had if I may ask:
I read through the paper on Probabilistic Matrix Factorization (Probabilistic Matrix Factorization) and noticed that simply by taking normal priors over the columns of the factor matrices (U and V), the Frobenius norms were directly penalized when considering the sum-of-squared errors objective function (eq. 4). Does this imply that I shouldn’t be explicitly penalizing the Frobenius norm of the matrix A if there is already a normal prior? Sorry for the unrelated question, but thought I’d ask and continue to pick your brain if you have the time.