Using Weighted Classes in CNN

In trying to set up my CNN layers with weighted classes I am getting an error that states:

"Error using classificationLayer>iParseInputArguments (line 58)

'ClassWeights' is not a recognized parameter. For a list of valid name-value pair arguments, see the

documentation for this function."

I am unsure if I'm using the class weights incorrectly. The following are the layers of my network:

classes = ["0","1"];

classWeights = 1./countcats(train_output);

classWeights = classWeights'/mean(classWeights);

layers = [

imageInputLayer([480 1 1])

convolution2dLayer([102 1],3,'Stride',1)

batchNormalizationLayer

leakyReluLayer

maxPooling2dLayer(2,'Stride',2,'Padding',[0 0 0 1])

convolution2dLayer([24 1],10,'Stride',1)

batchNormalizationLayer

leakyReluLayer

maxPooling2dLayer(2,'Stride',2,'Padding',[0 0 0 1])

convolution2dLayer([11 1],10,'Stride',1)

batchNormalizationLayer

leakyReluLayer

maxPooling2dLayer(2,'Stride',2,'Padding',[0 0 0 1])

convolution2dLayer([9 1],10,'Stride',1)

batchNormalizationLayer

leakyReluLayer

maxPooling2dLayer(2,'Stride',2,'Padding',[0 0 0 1])

fullyConnectedLayer(30)

fullyConnectedLayer(10)

fullyConnectedLayer(2)

softmaxLayer

classificationLayer('Classes',classes,'ClassWeights',classWeights)];

This error occurs with and without classes present in the brackets.

[Darrien Walters]

For further clarification i dont believe my layers are the real issue as i have run it before without the weighted classes and got no issue it occurs when i try to add class weights to the classification layer and i am unsure why if i am using the function wrong

Answer 1

[Rishik Ramena]

The 'ClassWeights' property for the classificationLayer has been introduced in MATLAB R2021a. So if you are running a version prior to this, you are supposed to get this error. Please find the documentation for your installed MATLAB version here.

[Darrien Walters]

Yes I wasnt aware of this at the time thank you for the response