Hyperparameter optimization fitrnet not working

Hi,

I'm trying to optimize the hyperparameters of a NN regression but I get an error regarding the inputs i give to the function fitrnet. I can't understand why since if I don't use ''OptimizeHyperparameters" but I specify the parameters my self the code works just fine.

%% training  Neural Network regression 
clear
clc
close all
load trainingSetReduced.mat 

test = struct2table(test);

predictorNames = {'v_i', 'E_tip', 'rho_tip', 'v_tip', 'Y_tip', 'Radius', 'E_plate', 'rho_plate', 'v_plate', 'Y_plate', 'Insulator','BC','anvil'}; 
predictors = test(:, predictorNames);
responseNames = {'F1','F2','F3','F4','F5','F6','F7','F8','F9','F10','F11','F12','F13','F14','F15','F16','F17','F18','F19','F20'};
response = test(:,responseNames);

X = table2array(predictors);
Y = table2array(response);



% Train the neural network
        regressionNeuralNetwork = fitrnet(...
        X, ...
        Y, ...
        'OptimizeHyperparameters', 'all', ...
        'HyperparameterOptimizationOptions', struct( ...
        'Optimizer', 'bayesopt', ...                          
        'AcquisitionFunctionName','expected-improvement-plus', ... 
        'UseParallel', true, ...                             
        'ShowPlots', true, ...                               
        'Verbose', 1, ...
        'MaxObjectiveEvaluations', 30));

Error using mlearnlib.internal.utils.ClassLabel (line 26)
You must pass class labels as a vector.

Error in classreg.learning.internal.ClassLabel

Error in classreg.learning.paramoptim.BayesoptInfo>numClasses (line 299)
N = numel(levels(classreg.learning.internal.ClassLabel(Y)));

Error in classreg.learning.paramoptim.BayesoptInfo/computeInMemoryDatasetStats (line 276)
this.NumClasses = numClasses(Response, ClassNamesPassed);

Error in classreg.learning.paramoptim.BayesoptInfo (line 243)
this = computeInMemoryDatasetStats(this, Predictors, Response, FitFunctionArgs, ObservationsInCols, IsRegression);

Error in classreg.learning.paramoptim.BayesoptInfoRNeuralNetwork (line 21)
[e-mail](Predictors, Response, FitFunctionArgs, false, false);

Error in classreg.learning.paramoptim.BayesoptInfo.makeBayesoptInfo (line 176)
Obj = ConstructorFcn(Predictors, Response, FitFunctionArgs);

Error in classreg.learning.paramoptim.fitoptimizing (line 30)
BOInfo = classreg.learning.paramoptim.BayesoptInfo.makeBayesoptInfo(FitFunctionName, Predictors, Response, FitFunctionArgs);

Error in fitrnet (line 148)
[this, varargout{1:nargout-1}] = classreg.learning.paramoptim.fitoptimizing("fitrnet",X,Y,varargin{:});

Does anybofy have some suggestions on how to solve the problem?

Thank you in advance

Answer 1

[the cyclist]

Per the documentation for fitrnet ...

"You can use an array ResponseVarName to specify multiple response variables. (since R2024b)"

but also

"Hyperparameter optimization options are not supported for multiresponse regression."

[Riccardo]

Thanks you! I completely missed that