train a deep network in matlab

main.m
% Import your data
[trainData, validationData, testData] = prepareData(); % Assume this function splits your data

% Define your architecture
numClasses = 2; % number of output classes
layers = [
    % Input layer
    imageInputLayer([28 28 1])

    % Hidden layers
    fullyConnectedLayer(512)
    reluLayer()
    fullyConnectedLayer(256)
    reluLayer()
    fullyConnectedLayer(128)
    reluLayer()

    % Output layer
    fullyConnectedLayer(numClasses)
    softmaxLayer()
    classificationLayer()
];

% Initialize your network
net = initializeNetwork(layers, trainData);

% Configure your training options
options = trainingOptions('adam', ...
    'MaxEpochs', 50, ...
    'MiniBatchSize', 128, ...
    'ValidationData', validationData, ...
    'ValidationFrequency', 10, ...
    'Plots', 'training-progress');

% Train your network
[net, info] = trainNetwork(trainData, layers, options);

% Evaluate your network
predictedLabels = classify(net, testData);
accuracy = sum(predictedLabels == testData.Labels)/numel(predictedLabels);
1037 chars
41 lines