1 function sheet11
   2 
   3 fprintf('loading data...');
   4 X = load('alpha_train_x.txt');
   5 Y = load('alpha_train_y.txt');
   6 
   7 [N, D] = size(X);
   8 
   9 fprintf('%d data points with %d features.\n', N, D);
  10 
  11 P = randperm(N);
  12 
  13 XE = X(P(N/2:end), :);
  14 YE = Y(P(N/2:end), :);
  15 X = X(P(1:N/2), :);
  16 Y = Y(P(1:N/2), :);
  17 
  18 size(X)
  19 size(Y)
  20 size(XE)
  21 size(YE)
  22 
  23 fprintf('training...');
  24 % find reasonable values below!
  25 epochs = ?;
  26 iters = ?;
  27 C = ?;
  28 eta = ?;
  29 C = linsvm_sgd(C, eta, epochs, iters, X, Y, XE, YE);
  30 
  31 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  32 %
  33 % your solution below.
  34 %
  35 %
  36 
  37 % Train a linear SVM with quadratic costs on the slack variables using
  38 % stochastic gradient descent.
  39 %
  40 % Parameters:
  41 %   - eta: base value for eta
  42 %   - epochs: total number of runs through the data.
  43 %   - iters: number of iterations per epoch
  44 %   - X, Y: training data
  45 %   - XE, YE: test data (only for computing the progress)
  46 %
  47 % Plot the test errors on training and test data for each epoch.
  48 function linsvm_sgd(C, eta, epochs, iters, X, Y, XE, YE)
  49 % ...