calculate_prediction_accuracy.m

用Cross validation的方法建立人工神经网络的模型！

function [accuracy]  = calculate_prediction_accuracy (sample1, features)

% THIS IS A SAMPLE FOR THIS FUNCTION
% MODIFY IT ACCORDING TO EACH PROBLEM SUCH THAT THE INPUTS/OUTPUTS ARE THE SAME
% FOR EXAMPLE, INSTEAD OF LOGISTIC REGRESSION USE NEURAL NETWORKS
% APPLY PCA IF NECESSARY
% ETC

% select features from the sample and add last column since this is the target variable
sample = sample1(:, [features size(sample1, 2)]);

nc = size(sample,2);
% Transform the target representation: class label '2/1' to '1/1'
i = find(sample(:, nc) == 2);
sample(i, nc) = 0;

% in 100 random selections estimate sensitivity and specificity of the predictor
% maybe 100 iterations is too many/few, consider changing it to some other value if necessary
for i = 1 : 2
    train = [];
    test = [];

    % find zero and one examples
    q0 = find(sample(:, size(sample, 2)) == 0);
    q1 = find(sample(:, size(sample, 2)) == 1);
    
    % randomly separate 10% from each into test set
    q0 = q0(randperm(length(q0)));
    q1 = q1(randperm(length(q1)));

    n0 = floor(0.1 * length(q0));
    test = [test; sample(q0(1 : n0), :)];
    n1 = floor(0.1 * length(q1));
    test = [test; sample(q1(1 : n1), :)];

    % make balanced training set
    train = sample(q1(n1 + 1 : length(q1)), :);
    train = [train; sample(q0(n0 + 1 : length(q0)), :)];
    
    % print out matrix dimensions and set some variables
    [n_examples, n_cols] = size(train);
    n_features = n_cols - 1;
    
    % normalize train and test separately
    [meanv, stdv, train(:, 1 : n_features)] = normalize(train(:, 1 : n_features), [], []);
    [meanv, stdv, test(:, 1 : n_features)] = normalize(test(:, 1 : n_features), meanv, stdv);
        
    % train predictor, ANN
    data = [train; test];
    [mse(i),R2(i),accuracy(i)] = classifierANN(data);
end

accuracy = mean(accuracy);

return