Hi all, I have a question about valid distributions in Stan with two parts: About fabs() Section 3.7. of the function reference manual lists Step-like functions with a warning saying that the functions listed there break gradients. I understand that is the case regarding all the functions except f…

Breaking gradients with fabs and the "double monomial"

rok_cesnovar July 20, 2021, 8:32am 2

fabs(x) is not differentiable at x = 0

1 Like

Topic		Replies	Views
Sqrt(square()) vs. fabs() General	6	2795	May 2, 2019
Yet another case of (Rejecting initial value: Gradient evaluated at the initial value is not finite.) Modeling	3	556	July 18, 2022
Initial value rejected, but the likelihood exists for that initial value Modeling fitting-issues , specification , cognitive-science	18	1311	July 13, 2020
Simple Example of an Improper Posterior Without Warnings from Stan General	22	2988	December 7, 2019
Integrate_1d in likelihood computation produces NaNs in the gradient Modeling integrator	18	1205	February 5, 2021