imds=imageDatastore( ...
    'SignsCutted', ...
    'IncludeSubfolders',true, ...
    'LabelSource','foldernames');

fprintf("Anzahl Bilder: %d\n", length(imds.Labels));

rng(7);
[imdsTrain, idmsValidation] = splitEachLabel(imds,0.7,'randomized');

fprintf("Trainingsmenge Anzahl Elemente: %d  Test(Validierungs)menge: %d\n", ...
    length(imdsTrain.Labels), length(idmsValidation.Labels));

net=alexnet;

layersTransfer = net.Layers(1:end-3);
numClasses = numel(categories(imdsTrain.Labels));

layers = [
    layersTransfer
    fullyConnectedLayer(numClasses, ...
        'WeightLearnRateFactor',20, ...
        'BiasLearnRateFactor',20)
    softmaxLayer
    classificationLayer
];

options = trainingOptions('sgdm', ...
    MiniBatchSize=10, ...                       % 10
    MaxEpochs=6, ...                            % 6
    InitialLearnRate=1e-4, ...                  % 1e-4
    ValidationData=idmsValidation, ...
    ValidationFrequency=3, ...                  % 3
    ValidationPatience=10, ...                  % 5
    Verbose=false, ...                          % false
    Plots='training-progress');

netTransfer = trainNetwork(imdsTrain, layers, options);

[yPred, scores] = classify(netTransfer, idmsValidation);

idx=randperm(numel(idmsValidation.Files), 64);

figure
for i=1 : 64
    subplot(8,8,i)
    I = readimage(idmsValidation, idx(i));
    imshow(I)
    label = yPred(idx(i));
    title(string(label));
end

accuracy = mean(yPred == idmsValidation.Labels);
fprintf("Genauigkeit: %3.2f\n", accuracy*100);