吴恩达机器学习第一次作业-线性回归

吴恩达机器学习第一次作业

(一)Linear regression with one variable

（1）Plot the data

addpath('你电脑上的数据文件路径')
data = load('ex1data1.txt')
 X = data(:,1);
 Y = data(:,2);
 plot(X,Y,'rx','MarkerSize',10);

（2）costFunction

function J = costFunction(X,y,theta)

m = length(y);
J = sum(((theta*X)-y).^2)/(2*m);

（3）Computing the cost J(θ)

m = length(y)
X = [ones(m,1),X]
theta =zeros(2,1);
J = costFunction(X,y,theta)  %回车我们可以得到J= 32.073

（4）Gradient Descent

alpha  = 0.01;
iterations = 1500;
%梯度下降函数为
function [theta,J_history] = gradientDescent(X,y,theta,alpha,num_iers)

m = length(y)
J_history = zeros(m,1);
for i=1:num_iers
	theta = theta-(alpha/m)*X'*(X*theta-y);
	J_history(i) = costFunction(X,y,theta);

end
end

%执行梯度下降函数

[theta,J_history] = gradientDescent(X,Y,theta,alpha,num_itera);
%得到theta为[ -3.6303;1.1664]，求得theta之后，我们就可以绘制出假设函数
plot(data(:,1),X*theta,'-');
predict1 = [1,3.5]*theta; 
%predict1 =  0.45198
predict2 = [1,7]*theta;

(二) Linear regression with multiple variables

（1）Feature Scaling

data = load('ex1data2.txt')
X = [data(:,1),data(:,2)]
y = [data(:,3)]

%featureScaling的函数如下
unction [X_norm,mu,sigma] = featureScaling(X,y)
X_norm = X;
mu = zeros(1,size(X,2))
sigma = zeros(1,size(X,2))
mu = mean(X_norm)
sigma = std(X_norm)
X_norm(:,1) = (X_norm(:,1)-mu(1))./sigma(1);
X_norm(:,2) = (X_norm(:,2)-mu(2))./sigma(2);
end

%调用featureSacling函数
[X_norm,mu,sigma] = featureScaling(X,y)
X = [ones(size(y),1),X_norm]
theta = zeros(3,1)
alpha = 0.01
iterations = 400

（2）Cost Function

function J = costFunctionOfMulti(X,y,theta)

 m = length(y);
 J = sum((X*theta-y).^2)/(2*m)

 %也可以写成J = (X*theta-y)'*(X*theta-y)/(2*m);

(3)Gradient Descent

function [theta,J_history] = GradientDescentOfMulti(X,y,theta,alpha,iterations)
m = length(y)
J_history = zeros(m,1)
for i= 1:iterations
	theta = theta - (alpha/m)*X'*(X*theta-y)
	J_history(i) = costFunctionOfMulti(X,y,theta)

end
end

% 调用GradientDescentOfMulti函数，其实单变量和多变量的代价函数和梯度下降
%可以相同
[theta,J_history] = GradientDescentOfMulti(X,y,theta,alpha,iterations)
%得到theta为
theta =
   334302.06399
   100087.11601
     3673.54845

(4)绘制迭代次数和代价之间的关系图

plot(1:numel(J_history),J_history,'-b','LineWidth',2);
![在这里插入图片描述](https://img-blog.csdnimg.cn/20191023095709370.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjQyNDI2OQ==,size_16,color_FFFFFF,t_70)
%改变alpha的值，重新调用梯度下降函数，重新绘制图像
alpha = 0.1
[theta,J_history] = GradientDescentOfMulti(X,y,theta,alpha,iterations)
plot(1:numel(J_history),J_history,'-b','LineWidth',2);![在这里插入图片描述](https://img-blog.csdnimg.cn/20191023095827694.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjQyNDI2OQ==,size_16,color_FFFFFF,t_70)

%我们可以看到当alpha变小时，函数收敛的速度变小。我们知道alpha代表的是学习效率
%如果我们把梯度下降的过程比作是下山的话，那么alpha就代表着我们迈步子的大小

（三）Normal Equation

（1）一次性求解theta的值

function [theta] = normalEqn(X, y)
theta = zeros(size(X, 2), 1);
theta = pinv(X'*X)*(X'*y);
end