Setting Borg parameters from the Matlab wrapper

In previous posts we talked about how to compile and run the Borg Matlab wrapper on Windows and OSX/Linux. To run the optimization after everything is compiled, the function call looks like this:

[vars, objs] = borg(11, 2, 0, @DTLZ2, 1000, zeros(1,11), ones(1,11), 0.01*ones(1,2), parameters);

In order, the arguments are:

- number of decision variables
- number of objectives
- number of constraints
- handle to function to be optimized
- number of function evaluations to perform
- decision lower bounds
- decision upper bounds
- epsilon values
- parameters

The last argument parameters is the topic of this post. This is a cell array that allows you to set the parameters without recompiling anything. For example, to set the random seed, you might do this:

parameters = {'rngstate', 12345, 'initialPopulationSize', 200, 'de.stepSize', 0.3}

before running the borg() function with parameters as the last argument. Notice this follows the Matlab cell array options format of {key, value, key, value ...}, so there must be an even number of items in the cell array.

Using the nativeborg.cpp file I tracked down all of the parameters (string keys) that the user can choose:

'rngstate': random seed
'initialPopulationSize'
'minimumPopulationSize'
'maximumPopulationSize'
'injectionRate'
'selectionRatio'
'maxMutationIndex'
'frequency': (for saving approximation sets during runtime)

// these two are enumerated types in the C code, not advised to change them
'restartMode'
'probabilityMode'

// operator-specific parameters
'pm.rate'
'pm.distributionIndex'
'sbx.rate'
'sbx.distributionIndex'
'de.crossoverRate'
'de.stepSize'
'um.rate'
'spx.parents'
'spx.epsilon'
'pcx.parents'
'pcx.eta'
'pcx.zeta'
'undx.parents'
'undx.zeta'
'undx.eta'

That’s all, thanks to Liz Houle at Boulder for the suggestion.

Advertisements