OK - in a quasi-related way, is it better in this case to encode the individual observations as bin-valued Categorical variables, rather than multinomials? It seems if I do this I can specify the observations from above as (zero-based):
[2, 2, 1]
Sorry that I’m asking about two different uses of the Categorical class - one to encode mixture components and the other for the Multinomial( n =1 ) case. Hopefully my question(s) make sense - let me know and I can clarify.