params = rand(3,3,4);
grad = ones(3,3,4);

averageSqGrad = [];

learnRate = 0.05;
sqGradDecay = 0.95;

[params,averageSqGrad] = rmspropupdate(params,grad,averageSqGrad,learnRate,sqGradDecay);

[XTrain,YTrain] = digitTrain4DArrayData;
classes = categories(YTrain);
numClasses = numel(classes);

layers = [
    imageInputLayer([28 28 1], 'Name','input','Mean',mean(XTrain,4))
    convolution2dLayer(5,20,'Name','conv1')
    reluLayer('Name', 'relu1')
    convolution2dLayer(3,20,'Padding',1,'Name','conv2')
    reluLayer('Name','relu2')
    convolution2dLayer(3,20,'Padding',1,'Name','conv3')
    reluLayer('Name','relu3')
    fullyConnectedLayer(numClasses,'Name','fc')];
lgraph = layerGraph(layers);

dlnet = dlnetwork(lgraph);

miniBatchSize = 128;
numEpochs = 20;
numObservations = numel(YTrain);
numIterationsPerEpoch = floor(numObservations./miniBatchSize);

executionEnvironment = "auto";

averageSqGrad = [];

plots = "training-progress";
if plots == "training-progress"
    iteration = 1;
    figure
    lineLossTrain = animatedline;
    xlabel("Total Iterations")
    ylabel("Loss")
end

for epoch = 1:numEpochs
    % Shuffle data.
    idx = randperm(numel(YTrain));
    XTrain = XTrain(:,:,:,idx);
    YTrain = YTrain(idx);
    
    for i = 1:numIterationsPerEpoch
        
        % Read mini-batch of data and convert the labels to dummy
        % variables.
        idx = (i-1)*miniBatchSize+1:i*miniBatchSize;
        X = XTrain(:,:,:,idx);
        
        Y = zeros(numClasses, miniBatchSize, 'single');
        for c = 1:numClasses
            Y(c,YTrain(idx)==classes(c)) = 1;
        end
        
        % Convert mini-batch of data to dlarray.
        dlX = dlarray(single(X),'SSCB');
        
        % If training on a GPU, then convert data to gpuArray.
        if (executionEnvironment == "auto" && canUseGPU) || executionEnvironment == "gpu"
            dlX = gpuArray(dlX);
        end
        
        % Evaluate the model gradients and loss using dlfeval and the
        % modelGradients function.
        [grad,loss] = dlfeval(@modelGradients,dlnet,dlX,Y);
        
        % Update the network parameters using the RMSProp optimizer.
        [dlnet,averageSqGrad] = rmspropupdate(dlnet,grad,averageSqGrad);
        
        % Display the training progress.
        if plots == "training-progress"
            addpoints(lineLossTrain,iteration,double(gather(extractdata(loss))))
            title("Loss During Training: Epoch - " + epoch + "; Iteration - " + i)
            drawnow
            iteration = iteration + 1;
        end
    end
end

[XTest, YTest] = digitTest4DArrayData;

dlXTest = dlarray(XTest,'SSCB');
if (executionEnvironment == "auto" && canUseGPU) || executionEnvironment == "gpu"
    dlXTest = gpuArray(dlXTest);
end

dlYPred = predict(dlnet,dlXTest);
[~,idx] = max(extractdata(dlYPred),[],1);
YPred = classes(idx);

accuracy = mean(YPred==YTest)

function [gradients,loss] = modelGradients(dlnet,dlX,Y)
    dlYPred = forward(dlnet,dlX);
    dlYPred = softmax(dlYPred);
    
    loss = crossentropy(dlYPred,Y);
    gradients = dlgradient(loss,dlnet.Learnables);
end