As a first step in building my model, I am trying to write a custom Dirichlet distribution and use it as a prior to estimate the parameters of a binomial distribution. The following clunky method works OK: from pymc3 import * from pymc3.distributions.multivariate import Dirichlet, Multinomial impor…

You dont need to create a theano tensor within the logp function, instead, rewrite it to take input, for example, something like: def dirich_logpdf(value): return -5.15209009879231 - 0.75 * tt.log(v).sum()

Can't vectorise custom distribution

Questions

Helmut May 24, 2018, 7:41am 3

Thanks. That has fixed it.

Topic		Replies	Views
Theano error with element wise multiplication Questions theano	5	2525	July 12, 2017
Fitting a distribution with custom functions Questions theano	5	3797	August 3, 2017
Running into Theano issues when using DensityDist Questions theano	6	1211	February 10, 2021
pm.Potential without Theano? Questions	14	2301	March 20, 2018
Integrating new distribution (wrapped c++ functions) - Problems with broadcasting and shapes Questions theano	3	715	July 8, 2019

Can't vectorise custom distribution

Related topics