Can I adjust patternsearch meshsize using output function (during optimisation)?

I am trying to adjust the 'meshsize' between certain iterations of a patternsearch optimisation. The output functions provides access to the patternsearch options, however this does not provide the variable meshsize. The meshsize is located in the optimvalues structure which is read-only (as far as I am aware). Is there any way to modify this without restarting the optimisation afresh?

For context, I am using a 'dynamic' cost function which changes slightly and I want to prevent premature convergence (and ensure a good search). If not, I will have to try and figure out a smooth way to restart optimisation without creating too much of a mess.

Thanks for any help!

p.s. I tried adjusting the MeshContractionFactor and MeshExpansionFactor to both be scalars giving the desired meshSize regardless of the next poll outcome. However, MeshContractionFactor has to be between 0 and 1. So that won't work.

[Robert Hamill]

@Alan Weiss Sorry if tagging is not encouraged but having read a lot of the optimisation questions, I thought you may well know the answer to this question.

Answer 1

[Alan Weiss]

Sorry to say, but a patternsearch output function does not allow you to change the mesh size. However, as I think you know, the output function does allow you to change options on the fly. But if you tried changing options already, which is what I think your comment on MeshContractionFactor is about, then I do not have any other ideas for you.

Alan Weiss

MATLAB mathematical toolbox documentation

[Robert Hamill]

Thanks Alan. Yes, I have decided to rework my approach to avoid resetting the mesh size.

For anyone reading in the future:

• You could manually set MeshContractionFactor and MeshExpansionFactor, and also temporaily set a 'fake' function value in order to force the desired change in mesh size.
• You could just restart the optimisation with a specified initial mesh size. (May be a pain in some cases).