Hello everyone, I am pretty new to Bayesian Inference so apologies in case this is a misunderstanding of the concepts. I am essentially trying to sample a variable “y” based on a Zero One Inflated Beta Regression model. Now, since this is zero one inflated beta regression, there are 3 distributions…

Are you trying to sample the variable y or is that a likelihood (the values are observed)? In the JAGS snippet it seems the observed data is just split into y.zero, y.one and y.c which you could do similarly in PyMC by adding three observed variables, two bernoulli and one beta, with the respective …

Yes, I am trying to sample the variable “y” and as per my understanding, I would have to use the equations representing the likelihood to do the sampling. I thought of splitting the data into multiple parts as you mentioned but wouldn’t that mean that I am just supplying an array of zeros, that of …

One would use numpy-like vectorization, instead of looping. You can look at linear regression examples in pymc-examples.

Yes, I am trying to sample the variable “y” and as per my understanding, I would have to use the equations representing the likelihood to do the sampling. Are you sampling y or is y your observed data?

I am sampling y_zero, y_one, and y_beta based on the observations from y. Sorry for the confusion.

Just to be clear if you have something like the following with pm.Model() as m: mu = ... sigma = ... y = pm.Normal("y", mu, sigma, observed=data) You wouldn’t be sampling y, it’s just a likelihood term used to infer the other unobserved paramerers that go into mu and sigma. y itself is obser…

Thanks for explaining this! I believe I misunderstood what was actually happening with the code. The samples should be drawn out for the parameters and not for y’s which are being used in the likelihood. I have updated my code to the following now, with pm.Model(): # Priors for unknow…

That could be precision issues. Try to parametrize the probabilities on the logit scale and pass logit_p directly to Bernoulli. You may also consider updating from pymc3 to newer versions of pymc > 5.0, as there have been many improvements in stability and model debugging.

Thanks for the help again! I tinkered with the data generated and changed the Bernoulli distributions to have logit_p parameterization. The sampling is working I believe but when I attempt to plot the trace using Arviz it complains that the sampling has only generated a single or a finite number of …

Zero One Inflated Beta Regression for an array of parameters

Questions

ricardoV94 June 21, 2023, 8:44pm 8

This might be relevant: Zero inflated Normal - #5 by ricardoV94

Topic		Replies	Views
Zero One Inflated Beta Regression Questions	14	1573	January 26, 2024
Contributing Zero-Inflated Beta Distribution Development	4	144	February 24, 2026
Literature on implementing a Zero-Inflated Beta likelihood Questions	19	2431	January 18, 2019
Zero and One inflated Beta Distribution Questions	5	1445	July 8, 2020
How to implement Zero-Inflated Multinomial? Questions	24	3075	November 10, 2019

Zero One Inflated Beta Regression for an array of parameters

Related topics