I am trying to define a model that looks like nvars, ncats = 3, 5 with pm.Model(): α = pm.Exponential('α', lam=λ, shape=(nvars, ncats)) b = pm.Dirichlet('b', a=α) trace = pm.sample() This fails, because the Dirichlet distribution expects an array, not a matrix, in PyMC3 (and in numpy,…

It should work, have you try specifying the shape of b?

Dynamic number of Dirichlet distributions?

colcarroll February 25, 2018, 12:35pm 3

I could have sworn I tried that… works fine once the shape is specified!

Once the shape pull request is settled, I might try to make sure that is detected automatically…

Topic		Replies	Views
Multivariate categorical with different probabilities Questions	5	1890	April 9, 2019
Shape issues when creating Dirichlet mixtures Questions	0	434	August 19, 2021
Latent Class Analysis Model (Dirichlet / Categorical Implementation Question) Questions	2	1220	February 10, 2019
Dirichlet and Categorical Questions	2	615	October 26, 2020
Dirichlet Distributions and Batching in pymc3 3.9+ Questions	2	470	February 1, 2021