function [netTransfer, info] = diatomnet(model_output_root, epoch_number, miniBatchSize, inputSize, numClasses, imdsTrain, imdsValidation, imdsTest)

augimdsTrain = augmentedImageDatastore(inputSize, imdsTrain);
augimdsValidation = augmentedImageDatastore(inputSize,imdsValidation);
augimdsTest = augmentedImageDatastore(inputSize,imdsTest);
   
lgraph = layerGraph();

tempLayers = [
    imageInputLayer(inputSize,"Name","data")
    convolution2dLayer([7 7],64,"Name","conv1-7x7_s2","BiasLearnRateFactor",2,"Padding","same","Stride",[2 2])
    reluLayer("Name","conv1-relu_7x7")
    maxPooling2dLayer([3 3],"Name","pool1-3x3_s2","Padding","same","Stride",[2 2])
    crossChannelNormalizationLayer(5,"Name","pool1-norm1","K",1)
    convolution2dLayer([1 1],64,"Name","conv2-3x3_reduce","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","conv2-relu_3x3_reduce")
    convolution2dLayer([3 3],192,"Name","conv2-3x3","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","conv2-relu_3x3")
    crossChannelNormalizationLayer(5,"Name","conv2-norm2","K",1)
    maxPooling2dLayer([3 3],"Name","pool2-3x3_s2","Padding","same","Stride",[2 2])];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],16,"Name","inception_3a-5x5_reduce","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","inception_3a-relu_5x5_reduce")
    convolution2dLayer([5 5],32,"Name","inception_3a-5x5","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","inception_3a-relu_5x5")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],96,"Name","inception_3a-3x3_reduce","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","inception_3a-relu_3x3_reduce")
    convolution2dLayer([3 3],128,"Name","inception_3a-3x3","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","inception_3a-relu_3x3")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],64,"Name","inception_3a-1x1","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","inception_3a-relu_1x1")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    maxPooling2dLayer([3 3],"Name","inception_3a-pool","Padding","same")
    convolution2dLayer([1 1],32,"Name","inception_3a-pool_proj","BiasLearnRateFactor",2,"Padding","same")
    reluLayer("Name","inception_3a-relu_pool_proj")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    depthConcatenationLayer(4,"Name","inception_3b-output")
    maxPooling2dLayer([3 3],"Name","pool3-3x3_s2","Padding","same","Stride",[2 2])];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],96,"Name","inception_4a-3x3_reduce","BiasLearnRateFactor",2)
    reluLayer("Name","inception_4a-relu_3x3_reduce")
    convolution2dLayer([3 3],208,"Name","inception_4a-3x3","BiasLearnRateFactor",2,"Padding",[1 1 1 1])
    reluLayer("Name","inception_4a-relu_3x3")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],16,"Name","inception_4a-5x5_reduce","BiasLearnRateFactor",2)
    reluLayer("Name","inception_4a-relu_5x5_reduce")
    convolution2dLayer([5 5],48,"Name","inception_4a-5x5","BiasLearnRateFactor",2,"Padding",[2 2 2 2])
    reluLayer("Name","inception_4a-relu_5x5")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],192,"Name","inception_4a-1x1","BiasLearnRateFactor",2)
    reluLayer("Name","inception_4a-relu_1x1")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    maxPooling2dLayer([3 3],"Name","inception_4a-pool","Padding",[1 1 1 1])
    convolution2dLayer([1 1],64,"Name","inception_4a-pool_proj","BiasLearnRateFactor",2)
    reluLayer("Name","inception_4a-relu_pool_proj")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    depthConcatenationLayer(4,"Name","inception_4e-output")
    maxPooling2dLayer([3 3],"Name","pool4-3x3_s2","Padding","same","Stride",[2 2])];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],48,"Name","inception_5b-5x5_reduce","BiasLearnRateFactor",2)
    reluLayer("Name","inception_5b-relu_5x5_reduce")
    convolution2dLayer([5 5],128,"Name","inception_5b-5x5","BiasLearnRateFactor",2,"Padding",[2 2 2 2])
    reluLayer("Name","inception_5b-relu_5x5")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    maxPooling2dLayer([3 3],"Name","inception_5b-pool","Padding",[1 1 1 1])
    convolution2dLayer([1 1],128,"Name","inception_5b-pool_proj","BiasLearnRateFactor",2)
    reluLayer("Name","inception_5b-relu_pool_proj")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],384,"Name","inception_5b-1x1","BiasLearnRateFactor",2)
    reluLayer("Name","inception_5b-relu_1x1")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    convolution2dLayer([1 1],192,"Name","inception_5b-3x3_reduce","BiasLearnRateFactor",2)
    reluLayer("Name","inception_5b-relu_3x3_reduce")
    convolution2dLayer([3 3],384,"Name","inception_5b-3x3","BiasLearnRateFactor",2,"Padding",[1 1 1 1])
    reluLayer("Name","inception_5b-relu_3x3")];
lgraph = addLayers(lgraph,tempLayers);

tempLayers = [
    depthConcatenationLayer(4,"Name","inception_5b-output")
    globalAveragePooling2dLayer("Name","pool5-7x7_s1")
    dropoutLayer(0.4,"Name","pool5-drop_7x7_s1")
    fullyConnectedLayer(numClasses,"Name","loss3-classifier","BiasLearnRateFactor",2)
    softmaxLayer("Name","prob")
    classificationLayer("Name","output")];
lgraph = addLayers(lgraph,tempLayers);

lgraph = connectLayers(lgraph,"pool2-3x3_s2","inception_3a-5x5_reduce");
lgraph = connectLayers(lgraph,"pool2-3x3_s2","inception_3a-3x3_reduce");
lgraph = connectLayers(lgraph,"pool2-3x3_s2","inception_3a-1x1");
lgraph = connectLayers(lgraph,"pool2-3x3_s2","inception_3a-pool");
lgraph = connectLayers(lgraph,"inception_3a-relu_1x1","inception_3b-output/in2");
lgraph = connectLayers(lgraph,"inception_3a-relu_pool_proj","inception_3b-output/in1");
lgraph = connectLayers(lgraph,"inception_3a-relu_5x5","inception_3b-output/in3");
lgraph = connectLayers(lgraph,"inception_3a-relu_3x3","inception_3b-output/in4");
lgraph = connectLayers(lgraph,"pool3-3x3_s2","inception_4a-3x3_reduce");
lgraph = connectLayers(lgraph,"pool3-3x3_s2","inception_4a-5x5_reduce");
lgraph = connectLayers(lgraph,"pool3-3x3_s2","inception_4a-1x1");
lgraph = connectLayers(lgraph,"pool3-3x3_s2","inception_4a-pool");
lgraph = connectLayers(lgraph,"inception_4a-relu_3x3","inception_4e-output/in4");
lgraph = connectLayers(lgraph,"inception_4a-relu_1x1","inception_4e-output/in3");
lgraph = connectLayers(lgraph,"inception_4a-relu_pool_proj","inception_4e-output/in1");
lgraph = connectLayers(lgraph,"inception_4a-relu_5x5","inception_4e-output/in2");
lgraph = connectLayers(lgraph,"pool4-3x3_s2","inception_5b-5x5_reduce");
lgraph = connectLayers(lgraph,"pool4-3x3_s2","inception_5b-pool");
lgraph = connectLayers(lgraph,"pool4-3x3_s2","inception_5b-1x1");
lgraph = connectLayers(lgraph,"pool4-3x3_s2","inception_5b-3x3_reduce");
lgraph = connectLayers(lgraph,"inception_5b-relu_1x1","inception_5b-output/in1");
lgraph = connectLayers(lgraph,"inception_5b-relu_pool_proj","inception_5b-output/in4");
lgraph = connectLayers(lgraph,"inception_5b-relu_5x5","inception_5b-output/in3");
lgraph = connectLayers(lgraph,"inception_5b-relu_3x3","inception_5b-output/in2");

clear tempLayers;

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

options = trainingOptions('sgdm', ...
    'MiniBatchSize',miniBatchSize, ...
    'MaxEpochs',1, ...
    'InitialLearnRate',0.0001, ...
    'Shuffle','every-epoch', ...
    'ValidationData',augimdsValidation, ...
    'ValidationFrequency',valFrequency, ...
    'Verbose',false, ...
    'Plots','none'...
);

fileID = fopen(strcat(model_output_root, '/results.txt'),'w');
fprintf(fileID,'epoch#;TrainingLoss;TrainingAccuracy;FinalValidationLoss;FinalValidationAccuracy;TrainAccuracy;TestAccuracy\n');  

 maxTestAccuracy = -1;

 for i = 1:epoch_number

    [netTransfer, info] = trainNetwork(augimdsTrain,lgraph,options);

    lgraph =  layerGraph(netTransfer);        

    % train
    [YPredTrain, scoresTrain] = classify(netTransfer,augimdsTrain);
    YTrain = imdsTrain.Labels;
    accuracyTrain = mean(YPredTrain == YTrain);       

    %test
    [YPredTest, scoresTest] = classify(netTransfer,augimdsTest);
    YTest = imdsTest.Labels;
    accuracyTest = mean(YPredTest == YTest);

    %validation
    [YPredValidation, scoresValidation] = classify(netTransfer,augimdsValidation);
    YVal = imdsValidation.Labels;
    accuracyValidation = mean(YPredValidation == YVal);
    
    sprintf('epoch#%d, train %3.4f, val %3.4f, test %3.4f, loss %3.4f'...
        ,i, accuracyTrain*100, info.FinalValidationAccuracy, accuracyTest*100, info.FinalValidationLoss)

     fprintf(fileID, 'epoch#%d;%3.4f;%3.4f;%3.4f;%3.4f;%3.4f;%3.4f\n'...
        ,i,...
        mean(info.TrainingLoss),...
        mean(info.TrainingAccuracy),...
        info.FinalValidationLoss,...      
        accuracyValidation,...
        accuracyTrain * 100,...
        accuracyTest * 100);      

    if accuracyTest > maxTestAccuracy
        %path = fullfile(model_output_root, strcat('model', int2str(i), '.mat'));
        %save (path, 'netTransfer', 'info', 'lgraph')
        maxTestAccuracy = accuracyTest;
        sprintf('saved  %d', i);
    end

    path = fullfile(model_output_root, strcat('model', int2str(i), '.mat'));
    save (path, 'netTransfer', 'info', 'lgraph')

 end

 fclose(fileID);

end