calculate the classification accuracy after training a "pretrained model"

how to calcualte the MSE, MAE RMSE or any other classification accuracy of a pretrained model?

next is my code:

imds = imageDatastore('C:\Users\Rayan\Desktop\Work\9_5_work_on_4_groups\9_1\R_9_1_GSM', ...

'IncludeSubfolders',true, ...

'LabelSource','foldernames');

[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomized');

numTrainImages = numel(imdsTrain.Labels);

idx = randperm(numTrainImages,16);

net = resnet50;

deepNetworkDesigner(net)

analyzeNetwork(net)

inputSize = net.Layers(1).InputSize;

lgraph = layerGraph(net);

edit(fullfile(matlabroot,'examples','nnet','main','findLayersToReplace.m'))

[learnableLayer,classLayer] = findLayersToReplace(lgraph);

[learnableLayer,classLayer] %#ok<NOPTS>

numClasses = numel(categories(imdsTrain.Labels));

%numClasses = 3

if isa(learnableLayer,'nnet.cnn.layer.FullyConnectedLayer')

newLearnableLayer = fullyConnectedLayer(numClasses, ...

'Name','new_fc', ...

'WeightLearnRateFactor',10, ...

'BiasLearnRateFactor',10);

elseif isa(learnableLayer,'nnet.cnn.layer.Convolution2DLayer')

newLearnableLayer = convolution2dLayer(1,numClasses, ...

'Name','new_conv', ...

'WeightLearnRateFactor',10, ...

'BiasLearnRateFactor',10);

end

lgraph = replaceLayer(lgraph,learnableLayer.Name,newLearnableLayer);

newClassLayer = classificationLayer('Name','new_classoutput');

lgraph = replaceLayer(lgraph,classLayer.Name,newClassLayer);

layers = lgraph.Layers;

connections = lgraph.Connections;

layers(1:20) = freezeWeights(layers(1:20));

lgraph = createLgraphUsingConnections(layers,connections);

augimdsTrain = augmentedImageDatastore(inputSize(1:2),imdsTrain)

augimdsValidation = augmentedImageDatastore(inputSize(1:2),imdsValidation);

miniBatchSize=10;

valFrequency = floor(numel(augimdsTrain.Files)/miniBatchSize);

options = trainingOptions('sgdm', ...

'MiniBatchSize',10, ...

'MaxEpochs',6, ...

'InitialLearnRate',0.0007, ...

'Shuffle','every-epoch', ...

'ValidationFrequency',valFrequency, ...

'ValidationData',augimdsValidation, ...

'Verbose',false, ...

'Plots','training-progress');

net = trainNetwork(augimdsTrain,lgraph,options);

[YPred,probs] = classify(net,augimdsValidation);

accuracy = mean(YPred == imdsValidation.Labels);

idx = randperm(numel(imdsValidation.Files),100);

R=1;

for j =1:24

figure(j)

for i = 1:4

subplot(2,2,i)

I = readimage(imdsValidation,idx(R));

imshow(I)

label = YPred(idx(R));

title(string(label) + ", " + num2str(100*max(probs(idx(R),:)),3) + "%");

R=R+1;

end

end

Answer 1

[Andreas Apostolatos]

Hi Rayan,

From the code snippet you share it appears that you are training a neural network for classification while you are then performing inference for some validation data,

net = trainNetwork(augimdsTrain,lgraph,options);
[YPred,probs] = classify(net,augimdsValidation);
accuracy = mean(YPred == imdsValidation.Labels);

Error measures such as the Mean Squarer Error (MSE) or the Root Mean Square Error (RMSE) are suited for regression problems where the response variables are continuous and not for classification problems.

To evaluate the performance of a classifier it is more appropriate to use a Confusion Matrix or to compute the percentage of responses that have been correctly predicted by the classifier. The corresponding workflow is underlined in the following link,

https://www.mathworks.com/help/stats/assess-neural-network-classifier-performance.html

I hope that you find this information useful for needs.

Kind regards

Andreas

[Rayan Matlob]

after training the previeus example , if i used (SVM) , would i be able to find MSE, MAE RMSE or any other classification accuracy

as follow:

% Call the workspace_5.mat and then run the following codes to achive the support vector machine

% Try SVM ############################################################################

[imdsTrain,imdsTest,augimdsValidation] = splitEachLabel(imds,0.7,0.2,0.1,'randomized');

numTrainImages = numel(imdsTrain.Labels);

%inputSize = net.Layers(1).InputSize;

augimdsTrain = augmentedImageDatastore(inputSize(1:2),imdsTrain);

augimdsTest = augmentedImageDatastore(inputSize(1:2),imdsTest);

augimdsValidation = augmentedImageDatastore(inputSize(1:2),imdsValidation);

layer = 'avg_pool';

featuresTrain = activations(net,augimdsTrain,layer,'OutputAs','rows');

featuresTest = activations(net,augimdsTest,layer,'OutputAs','rows');

whos featuresTrain

YTrain = imdsTrain.Labels;

YTest = imdsTest.Labels;

YVal = imdsValidation.Labels;

classifier = fitcecoc(featuresTrain,YTrain);

YPred = predict(classifier,featuresTest);

accuracy = mean(YPred == YTest)

layer = 'activation_46_relu';

featuresTrain = activations(net,augimdsTrain,layer);

featuresTest = activations(net,augimdsTest,layer);

whos featuresTrain

featuresTrain = squeeze(mean(featuresTrain,[1 2]))';

featuresTest = squeeze(mean(featuresTest,[1 2]))';

whos featuresTrain

classifier = fitcecoc(featuresTrain,YTrain);

YPred = predict(classifier,featuresTest);

accuracy = mean(YPred == YTest)

% with activation_46_relu, ---> accuracy=0.9846

augimdsValidation = augmentedImageDatastore(inputSize(1:2),imdsValidation);

[YPred,probs] = classify(net,augimdsValidation);

%Test

idx = randperm(numel(imdsValidation.Files),8);

R=1;

for j =1:2

figure(j)

for i = 1:4

subplot(2,2,i)

I = readimage(imdsValidation,idx(R));

imshow(I)

label = YPred(idx(R));

title(string(label) + ", " + num2str(100*max(probs(idx(R),:)),3) + "%");

R=R+1;

end

end

[Dehia]

Could you assist me in calculating the F-score, recall, sensitivity, and ROC curve, please?