吴恩达机器学习——exercise5

Regularized Linear Regression and Bias v.s. Variance

实现正则化线性回归模型，并且学习不同方差、偏差下的模型特性。

已有脚本：

ex5.m-Octave / MATLAB的引导脚本
ex5data1.mat-数据集
featureNormalize.m-归一化函数
fmincg.m-函数最小化例程（类似于fminunc）
plotFit.m-绘制多项式t
trainLinearReg.m-使用成本函数训练线性回归

编写脚本：

 linearRegCostFunction.m-正则化线性回归损失函数
 learningCurve.m-生成学习曲线
 polyFeatures.m-将数据映射到多项式特征空间
 validationCurve.m-生成交叉验证曲线

1、Regularized Linear Regression 正则化线性回归

1.1 Visualizing the dataset:

数据集分为三部分：

1. 训练集：X，y
2. 交叉验证集：Xval，yval
3. 测试集：Xtest，ytest

1.2 & 1.3 Regularized linear regression cost function: & Regularized linear regression gradient:

线性回归的损失函数（不要和logistic回归的损失函数记混了）：

线性回归的梯度计算公式：

linearRegCostFunction.m:

function [J, grad] = linearRegCostFunction(X, y, theta, lambda)
%LINEARREGCOSTFUNCTION Compute cost and gradient for regularized linear 
%regression with multiple variables
%   [J, grad] = LINEARREGCOSTFUNCTION(X, y, theta, lambda) computes the 
%   cost of using theta as the parameter for linear regression to fit the 
%   data points in X and y. Returns the cost in J and the gradient in grad

% Initialize some useful values
m = length(y); % number of training examples

% You need to return the following variables correctly 
J = 0;
grad = zeros(size(theta));

% ====================== YOUR CODE HERE ======================
% Instructions: Compute the cost and gradient of regularized linear 
%               regression for a particular choice of theta.
%
%               You should set J to the cost and grad to the gradient.
%
h = X * theta;
J = ((h-y)'*(h-y))/(2*m) + lambda*(theta(2:end,:)'*theta(2:end,:))/(2*m); %注意正则项是从j=1开始，而不是从j=0开始的
grad = ((h-y)'*X)/m + lambda*(theta')/m;

% =========================================================================

grad = grad(:);

end
 

注意：正则项是从第二项开始的，也就是下标为 j = 1 而不是 j = 0.如果正则项同样把第一项也算上，这个部分的损失函数相差不大，但是到后面的计算则会有较大的偏差。

2 Bias-variance: 偏差-方差

由于没有足够代表所有数据的特征的数据会导致模型有高偏差，也就是会使模型欠拟合；但是过多的特征又会使模型具有高方差，也就是趋于过拟合状态。

2.1 Learing curves：

学习曲线是检测学习算法的重要方式，其主要绘制了训练偏差和交叉验证偏差。

注意两个误差都是不包括正则项的；

learningCurve.m:

function [error_train, error_val] = ...
    learningCurve(X, y, Xval, yval, lambda)
%LEARNINGCURVE Generates the train and cross validation set errors needed 
%to plot a learning curve
%   [error_train, error_val] = ...
%       LEARNINGCURVE(X, y, Xval, yval, lambda) returns the train and
%       cross validation set errors for a learning curve. In particular, 
%       it returns two vectors of the same length - error_train and 
%       error_val. Then, error_train(i) contains the training error for
%       i examples (and similarly for error_val(i)).
%
%   In this function, you will compute the train and test errors for
%   dataset sizes from 1 up to m. In practice, when working with larger
%   datasets, you might want to do this in larger intervals.
%

% Number of training examples
m = size(X, 1);

% You need to return these values correctly
error_train = zeros(m, 1);
error_val   = zeros(m, 1);

% ====================== YOUR CODE HERE ======================
% Instructions: Fill in this function to return training errors in 
%               error_train and the cross validation errors in error_val. 
%               i.e., error_train(i) and 
%               error_val(i) should give you the errors
%               obtained after training on i examples.
%
% Note: You should evaluate the training error on the first i training
%       examples (i.e., X(1:i, :) and y(1:i)).
%
%       For the cross-validation error, you should instead evaluate on
%       the _entire_ cross validation set (Xval and yval).
%
% Note: If you are using your cost function (linearRegCostFunction)
%       to compute the training and cross validation error, you should 
%       call the function with the lambda argument set to 0. 
%       Do note that you will still need to use lambda when running
%       the training to obtain the theta parameters.
%
% Hint: You can loop over the examples with the following:
%
%       for i = 1:m
%           % Compute train/cross validation errors using training examples 
%           % X(1:i, :) and y(1:i), storing the result in 
%           % error_train(i) and error_val(i)
%           ....
%           
%       end
%

% ---------------------- Sample Solution ----------------------

for i =1:m
    X_temple = X(1:i,:);
    y_temple = y(1:i);
    theta = trainLinearReg(X_temple, y_temple, lambda);
    error_train(i,1) = ((X_temple*theta-y_temple)'*(X_temple*theta-y_temple)) / (2*m);
    error_val(i,1) = ((Xval*theta-yval)'*(Xval*theta-yval)) / (2*size(yval,1));
end

% -------------------------------------------------------------

% =========================================================================

end

3 Polynomial regression:

polyFeatures.m:

function [X_poly] = polyFeatures(X, p)
%POLYFEATURES Maps X (1D vector) into the p-th power
%   [X_poly] = POLYFEATURES(X, p) takes a data matrix X (size m x 1) and
%   maps each example into its polynomial features where
%   X_poly(i, :) = [X(i) X(i).^2 X(i).^3 ...  X(i).^p];
%

% You need to return the following variables correctly.
X_poly = zeros(numel(X), p);

% ====================== YOUR CODE HERE ======================
% Instructions: Given a vector X, return a matrix X_poly where the p-th 
%               column of X contains the values of X to the p-th power.
%
% 
for i = 1:p
     X_poly(:,i) = power(X,i);
end

% =========================================================================

end

3.3 Selecting λ using a cross validation set:

validationCurve.m:

function [lambda_vec, error_train, error_val] = ...
    validationCurve(X, y, Xval, yval)
%VALIDATIONCURVE Generate the train and validation errors needed to
%plot a validation curve that we can use to select lambda
%   [lambda_vec, error_train, error_val] = ...
%       VALIDATIONCURVE(X, y, Xval, yval) returns the train
%       and validation errors (in error_train, error_val)
%       for different values of lambda. You are given the training set (X,
%       y) and validation set (Xval, yval).
%

% Selected values of lambda (you should not change this)
lambda_vec = [0 0.001 0.003 0.01 0.03 0.1 0.3 1 3 10]';

% You need to return these variables correctly.
error_train = zeros(length(lambda_vec), 1);
error_val = zeros(length(lambda_vec), 1);

% ====================== YOUR CODE HERE ======================
% Instructions: Fill in this function to return training errors in 
%               error_train and the validation errors in error_val. The 
%               vector lambda_vec contains the different lambda parameters 
%               to use for each calculation of the errors, i.e, 
%               error_train(i), and error_val(i) should give 
%               you the errors obtained after training with 
%               lambda = lambda_vec(i)
%
% Note: You can loop over lambda_vec with the following:
%
%       for i = 1:length(lambda_vec)
%           lambda = lambda_vec(i);
%           % Compute train / val errors when training linear 
%           % regression with regularization parameter lambda
%           % You should store the result in error_train(i)
%           % and error_val(i)
%           ....
%           
%       end
%
%
for i = 1:length(lambda_vec)
    lambda = lambda_vec(i);
    theta = trainLinearReg(X, y, lambda);
    error_train(i) = linearRegCostFunction(X, y, theta, 0);
    error_val(i) = linearRegCostFunction(Xval, yval, theta, 0);
end
% =========================================================================

end

 

选作部分：

3.2 Computing test set error:

%% ===== Comuting test set error ========== 

lambda = 3;
theta = trainLinearReg([ones(size(X_poly,1),1) X_poly], y, lambda);
[J, grad] = linearRegCostFunction([ones(size(X_poly_test,1),1) X_poly_test],...
    ytest, theta, lambda);
fprintf('Cost J is ', J);
fprintf('grad is ', grad);

可是不知道为什么J = 13.20715，期望值是 J = 3.8599 ??？

希望知道的大佬不吝赐教