Hi, I was creating a logistic regression model and found that the observed 0/1 values didn’t match up with the probabilities which were supplied as parameters to a Bernoulli distribution. The code block below gives a minimal reproducible example: import numpy as np import pymc3 as pm import scipy …

Samples from prior appear to have wrong distribution

junpenglao October 7, 2018, 8:25am 2

This is likely related to https://github.com/pymc-devs/pymc3/issues/3210, which we are currently trying to fix.

Topic		Replies	Views
ValueError: probabilities are not non-negative when trying to sample prior predictive Questions	5	3611	September 24, 2020
New To PyMC3 \| Logistic Regression - Bug Questions bug	4	809	November 4, 2020
Model not reproducing observed data? Questions	9	1167	July 25, 2018
GLM logistic regression with custom prior in pymc3 (v. 3.6) Questions	10	4158	February 1, 2019
Hierarchical bernoulli model with bernoulli priors Questions	1	403	May 1, 2020