Improve error messaging at: Error evaluating initial BFGS point.

[jachymb]

This error doesn't really give any insight to the user as of what the actual problem with the init value is. Perhaps there is a nan or infinity in the loglikelihood? Maybe there's a problem with computing the gradients? Perhaps it's something else? Ideally it should give the user some insight how to fix the init value.

[bob-carpenter]

Hi, @jachymb and thanks for submitting.

Is this an issue with optimization or Pathfinder or both? Which interface is displaying this behavior?

Is the text of the exception not being printed to the console?

Would it be possible to supply a reproducible example in whatever interface you are using?

Thanks!

[jachymb]

I didn't save all the necessary things when I encountered this, but I will post it here if I see it again.

Clearly, this is raised at

stan/src/stan/optimization/bfgs.hpp

Line 157 in 30098c0

throw std::runtime_error("Error evaluating initial BFGS point.");

so I thought maybe simply more info could be added to the message that is available in the context. Is there is a perhaps detailed info about the type of error in the ret variable?

[bob-carpenter]

so I thought maybe simply more info could be added to the message that is available in the context.

I'm not sure we're getting useful return codes there. We can certainly say more about why an error is being thrown.