Get best hyperparameters for ionosphere data using fitcecoc & bayesopt. but i am getting error optimizableVariable.

Question

Vedant Chauhan 2022-3-17

评论： Vedant Chauhan 2022-3-24

load ionosphere

box = optimizableVariable('box',[1e-4,1e3],'Type','real','Transform','log');

kern = optimizableVariable('kern',[1e-4,1e3],'Type','real','Transform','log');

vars = [box,kern];

t = templateSVM('Standardize',true,'KernelFunction','gaussian','kernelscale',vars.kern ,'boxconstraint', vars.box);

fun = @(vars)kfoldLoss(fitcecoc(X,Y,'Learners',t,'FitPosterior',true,...

'ClassNames',{'g','b'},...

'Verbose',2,'Kfold',5));

results = bayesopt(fun,vars,'UseParallel',true);

The error coming is shown below:

No appropriate method, property, or field 'kern' for class 'optimizableVariable'.

Answer 1

Alan Weiss 2022-3-20

I do not understand why you are attempting to use fitcecoc with a 2-class model. Try this instead:

load ionosphere
mdl = fitcsvm(X,Y,"OptimizeHyperparameters","auto") % The auto parameters are box and kernel scale

Alan Weiss

MATLAB mathematical toolbox documentation

Vedant Chauhan 2022-3-24

Thank you