function [W, loss_log] = batchDescent(initial_W, dl, nc, svmbuff, ss, dataa, label, num_iters)
% data_length = 3072;
% num_classes = 10;
% svm_buffer = 1;
% step_size = 0.01;
data_length = dl;
num_classes = nc;
svm_buffer = svmbuff;
step_size = ss;
data = dataa;
labels = label;
num_iterations = num_iters;

%Might want to make these random values smaller
%W = rand(num_classes, data_length);
W = initial_W;

%Load CIFAR-10 data
%load('data_batch_1.mat');
batch_size = 32;
epochs = 100;
%num_imgs = length(data);
total_size = length(data);
num_imgs = batch_size;
loss_log = zeros(1,num_iterations);

for j = 1:num_iterations
    total_loss = 0;
    total_gradient = zeros(num_classes, data_length);
  
%for i = 1:num_imgs
for i = 1:epochs
    img_num = ceil(total_size.*rand(batch_size,1));
    %Calculate score of image i
    % img = im2double(data(i,:));
    for k=1:length(img_num)
    
        img = im2double(data(k,:));
        score_i = W*img';
        
        %Get the label of image i
        y = double(labels(k));
        
        tmp = score_i' - score_i(y+1) + svm_buffer;
        maxed = max(0,tmp);
        total_loss = total_loss + sum(maxed)-svm_buffer; %Subtract loss contributed by the correct image
        
        %Accumulate gradient
        bin_mask = find(maxed);
        bin_array = zeros(1,num_classes);
        bin_array(bin_mask) = 1;
        bin_array(y+1) = 0;
        total_gradient = total_gradient + bin_array' * img;
    end
end
    average_loss = total_loss / num_imgs;
    average_gradient = total_gradient / num_imgs;
    W = W - step_size * average_gradient;
    loss_log(j) = average_loss;
end
end