Hi Everyone, tl;dr Drawing from a Dirichlet distribution with shape (1,N): with pm.Model() as model: α = pm.Dirichlet(r'α', a=5*np.ones((1, N)), shape=(1, N)) appears to work in pyMC3 < 3.9 but fails in more recent versions with the error: Bad initial energy, check any log probabilities tha…

Dirichlet Distributions and Batching in pymc3 3.9+

junpenglao February 1, 2021, 8:36am 2

It seems to be a bug - could you raise an issue on Github?

Topic		Replies	Views
Dynamic number of Dirichlet distributions? Questions	3	622	February 25, 2018
Inferring Dirichlet concentration parameter Questions	11	1273	May 13, 2020
Large Feature Number Dirichlet Questions	6	1125	December 5, 2018
Multivariate categorical with different probabilities Questions	5	1890	April 9, 2019
How does pymc3 implement the Dirichlet distribution? Questions	7	2665	June 5, 2019