一、使用循环：
1.1原始版逻辑回归：
1 function g = sigmoid(z)
2 g = zeros(size(z));
3 g = 1 ./ (1 + exp(-z));
4 end
1 function [J, grad] = costFunction(theta, X, y)
2
3 % Initialize some useful values
4 m = length(y); % number of training examples
5
6 J = 0;
7 grad = zeros(size(theta));
8
9 n = size(theta);
10 gradtemp = 0;
11
12 %compute costfunction
13 for i = 1:m
14 temp1 = y(i) * log(sigmoid(X(i,:) * theta)) + (1 - y(i)) * log(1 - sigmoid(X(i,:) * theta));
15 J = J + temp1;
16 endfor
17
18 J = -(J / m);
19
20 %compute grade
21 for j = 1:n
22 for i = 1:m
23 temp2 = (sigmoid(X(i,:) * theta) - y(i)) * X(i,j);
24 gradtemp = gradtemp + temp2;
25 endfor
26 gradtemp = gradtemp / m;
27 grad(j) = gradtemp;
28 gradtemp = 0;
29 endfor
30
31 end
1 function p = predict(theta, X)
2
3 m = size(X, 1); % Number of training examples
4
5 p = zeros(m, 1);
6
7 for i =1:m
8 h = sigmoid(X(i,:) * theta);
9 if h >= 0.5
10 y = 1;
11 else
12 y = 0;
13 endif
14 p(i) = y;
15 endfor
16
17 end
1.2正则化版逻辑回归：
1 function [J, grad] = costFunctionReg(theta, X, y, lambda)
2
3 % Initialize some useful values
4 m = length(y); % number of training examples
5
6 J = 0;
7 grad = zeros(size(theta));
8
9 n = length(theta)
10 temp1 = 0;
11 temp2 = 0;
12 temp3 = 0;
13 result1 = 0;
14 result2 = 0;
15 %compute costfunction
16 for i = 1:m
17 h = sigmoid(X(i,:) * theta)
18 temp1 = y(i) * log(h) + (1 - y(i)) * log(1-h);
19 result1 = result1 + temp1;
20 endfor
21 for j = 2:n
22 temp2 = theta(j) * theta(j);
23 result2 = result2 + temp2;
24 endfor
25
26 J = (-1 / m) * result1 + lambda / (2 * m) * result2;
27
28 temp3 = 0
29 temp4 = 0
30 result3 = 0
31 for i = 1:m
32 h = (sigmoid(X(i,:) * theta))
33 temp3 = (h - y(i)) * X(i,1);
34 result3 = result3 + temp3;
35 endfor
36 grad(1) = (1 / m) * result3
37
38
39
40 %compute grade
41 for j = 2:n
42 result4 = 0;
43 for i = 1:m
44 h = (sigmoid(X(i,:) * theta))
45 temp4 = (h - y(i)) * X(i,j);
46 result4 = result4 + temp4;
47 endfor
48 grad(j) = (1 / m) * result4 + (lambda / m) * theta(j)
49 endfor
50
51 end
二、矩阵向量方式整体运算：
2.1原始版逻辑回归：
1 function g = sigmoid(z)
2
3 g = zeros(size(z));
4
5 g = 1 ./ (1 + exp(-z));
6
7 end
1 function [J, grad] = costFunction(theta, X, y)
2 % Initialize some useful values
3 m = length(y); % number of training examples
4
5 J = 0;
6 grad = zeros(size(theta));
7
8 h = sigmoid(X * theta);
9
10 J = sum(- y .* log(h) - (1 - y) .* log(1 - h)) / m;
11
12 grad = (1/m * sum((h - y).* X))';
13
14 end
1 function p = predict(theta, X)
2
3 m = size(X, 1); % Number of training examples
4
5 p = zeros(m, 1);
6
7 h = sigmoid(X * theta);
8
9 p(find(h >= 0.5)) = 1;
10
11 end
2.2正则化版逻辑回归：
1 function [J, grad] = costFunctionReg(theta, X, y, lambda)
2
3 % Initialize some useful values
4 m = length(y); % number of training examples
5
6 J = 0;
7 grad = zeros(size(theta));
8
9 n = size(theta,1);
10
11 h = sigmoid(X * theta);
12
13 J = sum(- y .* log(h) - (1 - y) .* log(1 - h)) / m + lambda / (2 * m) * sum(theta(2:n) .^ 2);
14
15 grad = (1/m * sum((h - y).* X))';
16
17 grad(2:n) = grad(2:n) + lambda / m * theta(2:n);
18
19 end