I am doing a generalized linear regression with Negative Binomial likelihood. This problem is different from a typical linear model where the x and y values are separate from each other. I am looking at timeseries data, where for each values in the series, I regress on the previous W values. I would…

My original solution wasn’t working because I had defined my distribution as a subclass of pm.Discrete, which defines univariate distributions. After reading some more of the documentation, I realized that the best way to do this was to define a custom likelihood function that took as arguments all …

Defining a custom multivariate distribution

Questions

junpenglao June 26, 2020, 5:58am 2

you mean NegativeBinomial only accept vector mu? I think we will need to modify its logp as well. What is the final shape of mu?

Topic		Replies	Views
Apply multivariate normal log-likelihood find the posterior distribution Questions	10	762	February 8, 2019
Likelihood given by product of two distributions? Questions	6	2215	August 4, 2018
How to set up a custom likelihood function for two variables Questions	8	13306	March 9, 2018
How to use theano to create a custom loglikelihood for an array of arrays Questions	13	2230	July 20, 2018
Issue with observed data in joint distribution Questions	3	984	May 15, 2018

Defining a custom multivariate distribution

Related topics