吴恩达机器学习---编程练习8

最新推荐文章于 2022-10-14 18:35:30 发布

DO-VIS

最新推荐文章于 2022-10-14 18:35:30 发布

阅读量215

点赞数 1

分类专栏： machine-learning 文章标签：机器学习

本文链接：https://blog.csdn.net/qq_37210730/article/details/99707881

版权

machine-learning 专栏收录该内容

16 篇文章 2 订阅

订阅专栏

博主只是初学机器学习的新人一枚，这篇博客旨在分享一下吴恩达机器学习课程编程练习8的答案，同时也是相当于自己对这一章的内容做一个回顾，让自己理解的更加的透彻，理性讨论，不喜勿喷

本章主题：Anomaly Detection and Recommender Systems(异常检测和推荐系统)

1.异常检测

假设特征服从正态分布，通过计算整体概率判断是否小于某个值，从而判断是不是正常的。
estimateGaussian.m

%estimate parameters
mu = sum(X) / m;
m_add = zeros(n-1,n);
mu_tmp = [mu;m_add];
mu_tmp = ones(m,n) * mu_tmp;
sigma2 = sum((X - mu_tmp).^2) / m;

selectThreshold.m

    yval_true = find(yval == 1);
    yval_false = find(yval == 0);
    p_true = find(pval <= epsilon);
    p_false = find(pval > epsilon);
    
    %actual class = 1  predicted class = 1
    truePositive = size(intersect(yval_true,p_true),1);
    %actual class = 0  predicted class = 1
    falsePositive = size(intersect(yval_false,p_true),1);
    %actual class = 1  predicted class = 0
    falseNegative = size(intersect(yval_true,p_false),1);

    precision = truePositive / (truePositive + falsePositive);
    recall = truePositive / (truePositive + falseNegative);

    F1 = 2*precision*recall / (precision + recall);

2.推荐系统

根据已有的信息来进行预测，有点类似于线性回归（或者说叫多元线性回归？）
cofiCostFunc.m

Program 1----for-loop

J =  0.5 * sum(sum(R.*((X*Theta'-Y).^2))) + (lambda/2)*(sum(sum(Theta.^2)) + sum(sum(X.^2))); 

for i = 1:num_movies
    idx = find(R(i,:) == 1);
    Theta_temp = Theta(idx,:);
    Y_temp = Y(i,idx);
    A = X(i,:);
    t = (X(i,:) * Theta_temp' - Y_temp);
    X_grad(i,:) = (X(i,:) * Theta_temp' - Y_temp) * Theta_temp + lambda*X(i,:); 
end

for i = 1:num_users
    idx = find(R(:,i) == 1);  %every for all movies
    Theta_grad(i,:) = (X(idx,:)*Theta(i,:)' - Y(idx,i))' * X(idx,:) + lambda*Theta(i,:);
end

Program 2----vectorized

J = (1 / 2) * sum(sum(((X * Theta' - Y) .* R) .^ 2));
J = J + (lambda / 2) * sum(sum(Theta .^ 2)) + ...
(lambda / 2) * sum(sum(X .^ 2));

X_grad = ((X * Theta' - Y) .* R) * Theta + lambda * X;
Theta_grad = ((X * Theta' - Y) .* R)' * X +  lambda * Theta;