Mixture of hierarchical model

[peppe]

Hi thanks for the quick replies.

I tried to add this:

Mixture of hierarchical model

model 1 is definitely in the right direction. However, you should still try to restrict the coefficient to make sure there is no mode switching. As a quick fix try adding this constraint into your model1: # break symmetry order_means_potential = pm.Potential('order_means_potential', tt.switch(mu_b[1]-mu_b[0] < 0, -np.inf, 0))

but I got

ValueError: Bad initial energy: inf. The model might be misspecified.

With respect to your second suggestion

Mixture of hierarchical model

Also, you might find some inspiration from this question: I had two notebook that explore different parameterization and inference:

I am not sure where I should get inspiration from. Are you suggesting to “switch label by hand”?