I am using Julia BlackBoxOptim to search for global optima of a function, using default optimizer. The fitness value keeps unchanged for 150 evaluations, so I am wondering whether it is possible to have an option similar to f_tol in optim.jl? That way the solver can be considered to converge when the decrease in fitness value is less than a specific tolerance level?
I checked fixed shutdown methods and added test shutting down optimization from… · robertfeldt/BlackBoxOptim.jl@5ed8cd9 · GitHub, but still do not have an idea of how to do that.
That would be better than MaxStep and MaxTime because different settings of functions have very different number of evaluations and time for convergence and some requires a large MaxTime which will be a waste of time for others.
A part of the log file is shown as below:
65218.22 secs, 4965 evals, 4888 steps, improv/step: 0.208 (last = 1.0000), fitness=1826.037268100 65230.30 secs, 4966 evals, 4889 steps, improv/step: 0.208 (last = 0.0000), fitness=1826.037268100 65242.76 secs, 4967 evals, 4890 steps, improv/step: 0.208 (last = 0.0000), fitness=1826.037268100 ...... 67005.57 secs, 5116 evals, 5039 steps, improv/step: 0.207 (last = 0.0000), fitness=1826.037268100 67017.29 secs, 5117 evals, 5040 steps, improv/step: 0.207 (last = 1.0000), fitness=1826.037268100
Thank you in advance!