Hi @Gon_F, sorry for not replying earlier. I was a bit busy to look much at the discourse. From what I see, your problem is that price will end up having (N, 1) shape, instead of just (N,). Could you try setting beta's shape to (270, 1). If not, you can also try to tt.squeeze(price) when you pass it as y_lik's mu.
I think that the true origin of the problem is that when you set observed=np.log(SalePrice), the sale’s price shape will be used to determine the y_lik variable’s shape. Currently in pymc3, distribution parameter values are not taken into account for the determination of shape, and will likely remain that way. pymc4 will not be like this because it will work with tensorflow distribution’s which are handled as tensors themselves, with some added flavors.
Nevertheless, to tell you a more precise answer, I would need to know the full traceback of the exception, which looks like it is raised from distributions.generate_samples.