diff: Return number of function evaluations from finite difference functions?

Question

diff: Return number of function evaluations from finite difference functions?

Closed this issue 7 years ago · 3 comments

Answer 1 · 2016-03-25T14:09:09.000Z

I don't think that's necessary. It's very predictable, isn't it? Are you thinking about adding to the number of calls in optimize?

Answer 2 · 2016-03-25T14:34:48.000Z

Yes, it's kind of predictable, the number of evaluations implicitly follows from the method. I think that we should explicitly write the number in the docs. I wanted to know your opinion.

And yes, I was having optimize in mind even though I think it would be nice if we didn't mix the stats from the method itself and from fd approximations. Speaking of optimize, we will need to provide a mechanism for calling Gradient and/or Hessian that sets OriginKnown and Value.

Answer 3 · 2017-08-03T17:23:53.000Z

If we want to support this, I think it should be implemented with functions. For example, func HessianEvals(stencil Stencil, originKnown bool) int.

Feel free to re-open this issue in the main repository if you think it's still a good idea.