TargetFitness?

[MA-Kochen]

Hello,

I am optimizing systems of differential equations (in the form of simulatable SBML models). The objective function takes the difference between a set of data and the simulated trace for the species in the model. Setting TargetFitness seems to have no effect. When running something like

bboptimize(objective; SearchRange=sample_distributions, PopulationSize=50, TargetFitness=100.0)

it blows through the target fitness and runs until the max step limit is reached. The output looks something like

Optimization stopped after 10001 steps and 1.47 seconds
Termination reason: Max number of steps (10000) reached
Steps per second = 6803.36
Function evals per second = 6885.00
Improvements/step = 0.23490
Total function evaluations = 10121
Best candidate found: [-2.17269e-8, -2.0, -0.522879]
Fitness: 0.000000001

Am I missing something here?

[robertfeldt]

Thanks for reporting this. It does seem like this is not properly implemented, yes. Should be a big thing though. So the functionality you would expect is that if it reaches a fitness <= the TargetFitness the optimization terminates and the result is returned? I need to check how other optimization packages does this so we can try to stay close in naming and behavior.

[robertfeldt]

"Shouldn't be a big thing though", is what I meant. ;)

[robertfeldt]

Actually, there is a way you can do this yourself via a callback function. Please see the test code from line 239 in this file for an example:

https://github.com/robertfeldt/BlackBoxOptim.jl/blob/master/test/test_toplevel_bboptimize.jl

[MA-Kochen]

"So the functionality you would expect is that if it reaches a fitness <= the TargetFitness the optimization terminates and the result is returned? " Yes, that is what I would expect. I'll take a look at that test code.

[MA-Kochen]

Can confirm that the callback function works. Thanks!

[robertfeldt]

Great. Note that you don't need to set CallbackInterval to zero (since it is a bit inefficient since it then call the callback every step of the optimization. If you set it to 1.0 or so it will stop the optimization within one second of reaching the target etc.