Perform model fit evaluation in Bayesian way when sampling from the custom distribution is not known

ricardoV94 · October 20, 2023, 12:33pm

I’ve implemented the base InverseGaussian in this gist, and then a shifted version:

gist.github.com

https://gist.github.com/ricardoV94/77751556f7df06503d72b93634671e32

Shifted Wald distribution.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "db6ad00c",
   "metadata": {},
   "source": [
    "# Implementing a Shifted Wald Distribution in PyMC"
   ]
  },

This file has been truncated. show original

Adding the logcdf function to the InverseGaussian should allow the shifted version to work with pm.Censored, but for the point of doing prior and posterior predictive I assume you don’t want to censor the values anyway, so you can just write a model with the “uncensored” version to do predictions?

Topic		Replies	Views
Sample posterior predictive error Questions	3	1209	October 8, 2019
Using CustomDist for sample_prior_predictive v5 prior , modeling , pytensor , diagnostics	9	540	July 12, 2023
Simulating Fake Data from pymc3 model Questions	1	1783	October 28, 2017
Sample from prior? Questions	9	3525	July 20, 2018
Having problems defining random function for custom distribution with mixture of observed and random variables Questions	5	1045	November 20, 2018

Perform model fit evaluation in Bayesian way when sampling from the custom distribution is not known

Related topics