首页 > 代码库 > Logistic Regression & Regularization ----- Stanford Machine Learning(by Andrew NG)Course Notes

Logistic Regression & Regularization ----- Stanford Machine Learning(by Andrew NG)Course Notes

coursera上面Andrew NG的Machine learning课程地址为:https://www.coursera.org/course/ml

我曾经使用Logistic Regression方法进行ctr的预测工作,因为当时主要使用的是成型的工具,对该算法本身并没有什么比较深入的认识,不过可以客观的感受到Logistic Regression的商用价值。

Logistic Regression Model

A. objective function

      其中z的定义域是(-INF,+INF),值域是[0,1]

We call this function sigmoid function or logistic function.

We want 0 ≤ hθ(x) ≤ 1   and  hθ(x) = g(θTx)  

B. Decision boundary

  在 0  hθ(x)  1的连续空间内,用logistic regression做分类时,我们可以将hθ(x)等于0.5作为分割点。

   if  hθ(x) ≥ 0.5,predict "y = 1";

    if  hθ(x) < 0.5,predict "y = 0";

 而Decision Boundary就是能够将所有数据点进行很好地分类的 h(x) 边界。

C. Cost Function

Defination

 

Because y = 0 or y = 1,and cost function can been writen as below:

 


Advanced optimization

In order to minimize J(θ),  and get θ. Then how to get minθ J(θ) ?

A. Using gradient descent to do optimization

 Repeat{

            }            

 Compute , we can get (推导过程下方附录)

 Repeat{

            }

B.其他基于梯度的优化方法

   a) Conjugate gradient(共轭梯度)

  b) 牛顿法

  c)  拟牛顿法

       d) BFGS(以其发明者Broyden, Fletcher, Goldfarb和Shanno的姓氏首字母命名),公式:

  e) L-BFGS

  f)  OWLQN


Multi classification

How to do multi classification using logistic regression?   (one vs rest)

A. How to train model?

     当训练语料标注的类别大于2时,记为n。我们可以训练n个LR模型,每个模型的训练数据正例是第i类的样本,反例是剩余样本。(1≤ i ≤n)

 

B.How to do prediction? 

    在 n 个 hθ(x) 中,获得最大 hθ(x) 的类就是x所分到的类,即 


Overfitting

A. How to address overfitting?

     a) Reduce number of features.

            ― Manually select which features to keep.

            ― Model selection algorithm (later in course).

      b) Regularization(规范化)

            ― Keep all the features, but reduce magnitude/values of all parameters .

            ― Works well when we have a lot of features, each of which contributes a bit to predicting .

      c) Cross-validation(交叉验证)

            ― Holdout验证: 我们将语料库分成:训练集,验证集和测试集;

            ― K-fold cross-validation:优势在于同时重复运用随机产生的子样本进行训练和验证,每次的结果验证一次;

B. Regularized linear regression

   

                (式1)

         (式2)

C. Normal equation

    Non-invertibility(optional/advanced).

    suppose m ≤ n                 m: the number of examples;    n: the number of features;

                θ = (XTX)-1XTy

由(式1)和(式2)可以得到对应的n+1维参数矩阵。

D. Regularized logistic regression

    Regularized cost function:

     J(θ) =   

   Gradient descent: 

      Repeat{

                

              } 


Logistic Regression与Linear Regression的关系

   Logistic Regression是线性回归的一种,Logistic Regression 就是一个被logistic方程归一化后的线性回归。


Logistic Regression的适用性

  1) 可用于概率预测,也可用于分类;

  2) 仅能用于线性问题;

  3) 各feature之间不需要满足条件独立假设,但各个feature的贡献是独立计算的。


 HOMEWORK

好了,既然看完了视频课程,就来做一下作业吧,下面是Logistic Regression部分作业的核心代码:

1.sigmoid.m

m = 0;
n=0;
[m,n] = size(z);
for i = 1:m
   for j = 1:n
      g(i,j) = 1/(1+e^(-z(i,j)));
   end
end

2.costFunction.m

for i =1:m
   J = J+(-y(i)*log(sigmoid(X(i,:)*theta)))-(1-y(i))*log(1-sigmoid(X(i,:)*theta));
end
J=J/m;
for j=1:size(theta)
   for i=1:m
      grad(j)=grad(j)+(sigmoid(X(i,:)*theta)-y(i))*X(i,j);
   end
grad(j)=grad(j)/m;
end

3.predict.m

for i=1:m
   if(sigmoid(theta‘*X(i,:)‘)>0.5)
      p(i)=1;
   else
      p(i)=0;
   endif
end

4.costFunctionReg.m

for i =1:m
   J = J+(-y(i)*log(sigmoid(X(i,:)*theta)))-(1-y(i))*log(1-sigmoid(X(i,:)*theta));
end
J=J/m;
for j=2:size(theta)
    J = J+(lambda*(theta(j)^2)/(2*m));
end

for j=1:size(theta)
   for i=1:m
      grad(j)=grad(j)+(sigmoid(X(i,:)*theta)-y(i))*X(i,j);
   end
grad(j)=grad(j)/m;
end
for j=2:size(theta)
   grad(j)=grad(j)+(lambda*theta(j))/m;
end

 附录

Logistic regression gradient descent 推导过程