search

bbopt / NOMAD.jl

Julia interface to the NOMAD blackbox optimization software
Other
47 stars 5 forks source link

Relationship between the granularity and the min_mesh_size of a NomadProblem #58

Closed MonssafToukal closed 1 year ago

MonssafToukal commented 2 years ago

I would like to know if there is a relationship between the granularity and the min_mesh_size of a NomadProblem. I am currently trying to increase the granularity of real variables, but if I do so, should I also change the mesh size min_mesh_size?

salomonl commented 2 years ago

The granularity is always bigger or equal to the min_mesh_size. Another point is that if if you reach the target granularity, Nomad will only stop after consuming the whole budget of evaluations. If you want to stop it faster, I think it is better to tune the min_mesh_size parameter than the granularity parameter. If you have integer or binary variables, it is better to declare their types than to set the granularity.

Note also that the min_mesh_size parameter can never be inferior to 1e-13 (this behavior could be changed in the future).