I'm trying to apply conditional constraints on two variable with bayesopt

here are part the paramters i'm trying to optimize

optimVars = [optimizableVariable('numlayers',[20 25],'Type','integer')

optimizableVariable('LR',[1e-3 1e-1],'Transform','log')

optimizableVariable('MBS',[20 32],'Type','integer')

optimizableVariable('RP',[1e-3 1e-1],'Transform','log')];

function Xnew = condvariablefcn(X)

Xnew=X;

Xnew.L2Regularization(Xnew.numLayersToFreeze==24) = NaN;

end

but when i run the code, i get this message :

Error using BayesianOptimization/applyXConstraint

The XConstraintFcn must return a logical column vector the same height as

its input (10000), but it returned an object of class table with size

[10000 4].

the code runs fine before i added this constraint. so what is the problem with the function?

Answer 1

[Alan Weiss]

I think that you set the wrong name-value argument for your constraint function. bayesopt thinks that you are using an XConstraint, not a conditional constraint. In your bayesopt call set the name-value argument like this:

results = bayesopt(fun,vars,'ConditionalVariableFcn',@condvariablefcn)

Alan Weiss

MATLAB mathematical toolbox documentation