How can we build a mixture of mixtures?

It should works also for B version with the shape specified for the Mixture likelihood:
ln_mix = pm.Mixture.dist(w=ln_w,comp_dists=ln_components, shape=x.shape)