gradient descent

%data = csvread('train.csv');

trainingSetSize = 250;

for i = 1:trainingSetSize
    img = data(i, 2:end);
    for j = 1:length(img)
        if img(j) > 0
            img(j) = 1;
        end
    end
    data(i, 2:end) = img;
end

function v = adjustWeights(w, trainingSet, learningRate)
    dw = zeros(1, length(w));
    for i = 1:size(trainingSet, 1)
        t = trainingSet(i, 1);
        x = trainingSet(i, 2:end);
        dw = dw + (dot(w, x) - t) * x;
    end
    v = w - learningRate * dw;
end

w = rand(1, 28*28);
learningRate = 0.001;
for i = 1:10
    w = adjustWeights(w, data(1:trainingSetSize, :), learningRate / i);
end

w