I am Designging a multi-ouput ANN for regression and classification simultaneously but i face error

clc;clear all;
%% Read file
filename = "CorrectDS.csv";
tbl = readtable(filename,'TextType','String');
head(tbl)
X=table2array(tbl);
training=X(:,1:23);
classLabel=categorical(X(:,24));
regLabel=X(:,25);
num_features = size(X, 2)-2;
num_classes = 2;
num_neurons_hidden = 20;
% Define the neural network architecture
layers = [
    featureInputLayer(num_features, 'Normalization', 'none')
    fullyConnectedLayer(num_neurons_hidden)
    reluLayer
    fullyConnectedLayer(num_neurons_hidden)
    reluLayer
    fullyConnectedLayer(num_neurons_hidden)  % Additional hidden layer for classification
    reluLayer
    fullyConnectedLayer(num_classes)        % Classification output
    softmaxLayer
    fullyConnectedLayer(1)                  % Regression output
    regressionLayer
    ];
options = trainingOptions('adam', ...
    'MaxEpochs', 50, ...
    'MiniBatchSize', 16, ...
    'Verbose', true, ...
    'Plots', 'training-progress');
lab={classLabel,regLabel};
net = trainNetwork(training,lab,layers, options);

and I got this error

Error using trainNetwork
Invalid training data. For classification tasks, responses must be a vector of categorical responses. For regression tasks, responses must be a vector, a matrix,
or a 4-D array of numeric responses which must not contain NaNs.

Error in MAin (line 40)
net = trainNetwork(training,lab,layers, options);

Please, Someone help me to tackle this error

[Walter Roberson]

classLabel=categorical(X(:,24));
%...
lab={classLabel,regLabel};
net = trainNetwork(training,lab,layers, options);

Your lab entry is a cell array in which the first entry is categorical (and the second entry is probably numeric)

Invalid training data. For classification tasks, responses must be a vector of categorical responses. For regression tasks, responses must be a vector, a matrix,

or a 4-D array of numeric responses which must not contain NaNs.

Cell array does not match any of those possibilties.

[Ameer HAmza]

So how i can solve it?

[Walter Roberson]

classLabel = X(:,24);

lab = [classlabel, regLabel];

[Ameer HAmza]

ok thank you soo much

[Walter Roberson]

You probably cannot do regression and classification simultaneously.

[Ameer HAmza]

is there any other way to do this?

[Walter Roberson]

You probably have to split the regression and the classification.

[Ameer HAmza]

ok thank you

Answer 1

[Kaustab Pal]

Hi @Ameer HAmza,

To create a Neural Network capable of both classification and regression, you should design the network with two distinct branches: one dedicated to classification and the other to regression. You can utiize the "dlnetwork" function to achieve this. Below is a sample code to illustrate how you can implement this:

num_classes = 3;
num_neurons_hidden = 20;
% Define the neural network architecture
net = dlnetwork();
tmp_layers = [
    featureInputLayer(10, 'Normalization', 'none', "Name","input_layer")
    fullyConnectedLayer(num_neurons_hidden,"Name","fc1")
    reluLayer("Name","relu_1")
    fullyConnectedLayer(num_neurons_hidden, "Name","fc2")
    reluLayer("Name","relu_2")
    fullyConnectedLayer(num_neurons_hidden, "Name","fc3")  % Additional hidden layer for classification
    reluLayer("Name","relu_3")
    fullyConnectedLayer(num_classes, "Name","fc4")        % Classification output
    softmaxLayer("Name","classifier_output")
    ];
net = addLayers(net,tmp_layers);
tempLayers = [
    fullyConnectedLayer(1, "Name","regression_output") 
    ];
net = addLayers(net,tempLayers);
net = connectLayers(net,"fc4","regression_output");

analyzeNetwork(net)

The output of the analyzeNetwork will look something like this:

You can now incorporate the appropriate loss functions for each branch to train the network on your dataset effectively.

To know more about dlnetwork, you can refer the official documentation here: https://www.mathworks.com/help/deeplearning/ref/dlnetwork.html

I hope this helps!

Answer 2

[Ameer HAmza]

Thank you brother