Debug help

Modeling
1

I have a very complex model with 100+ lines of code (which I cannot share - sorry), I do get this error below

“Rejecting initial value:”
[299] " Gradient evaluated at the initial value is not finite."
[300] " Stan can’t start sampling from this initial value."
[301] "Initialization between (-2, 2) failed after 100 attempts. "

I cannot see any parameters that are bound between -2 and 2. Is there a way for stan to be more explicit about the affected parameter?

Dirk

2

The -2 to 2 thing is on the unconstrained space; specify the init_r argument to some value less than 2 to use a smaller window. But, no, Stan cannot be made more explicit about the affected parameter because the initialization problem is related to the joint distribution.

3

You should be getting error messages that tell you what got rejected if you’re using a recent version of Stan.