Watermelon Book + pumpkin book chapter 1-2

1. Watermelon Chapter 1

1.1 Basic terminology

1.2 After-school exercises

I feel like I didn't do anything right , Not finished , Still need to continue thinking

2. Watermelon Book II

2.1 Basic terminology

2.2 Evaluation methods

2.2.1 Set aside method

2.2.2 Cross validation

2.2.3 Self help law

Every time from the dataset D Extract a data from , Hypothesis extraction m Time , Extraction of m The data obtained at this time forms a data set A; The remaining data is probably 0.368 Forming data sets B.
shortcoming ： The self-help method does not have stratified sampling , Changed the data distribution , It introduces estimation bias .

2.3 Performance estimation

2.3.1 Precision and recall

Precision rate ：P
p = The prediction is positive and correct /（ All that are predicted to be positive ）
Recall rate ：R
R = The prediction is positive and correct / （ The number of positive samples in the original data set ）

2.4 After-school exercises

3. Pumpkin book chapter 1

3.1 The formula 1.1 understand

Purpose ： Calculate the samples other than the training set in the learner

The error of the .( Under the learner A Express )
computing method ：

sum = 0
for(i = h) {
         //h It is the hypothesis generated by the learner , Go through all the assumptions 
	for(j = f(x)) {
        // All over x,x It belongs to data other than training set data 
		sum += (h(x) != f(x) ? p(x) : 0) * p(h|X, A); //p(x)
	}
}

h It's a learning device A Through samples X The resulting assumptions ; It is worth noting that h There are many possibilities .
therefore sigma P(h | X, εa) = 1

3.2 The formula 1.2 understand

understand ： Formula one considers only one objective function , Formula 2 considers many, many objective functions ;（ I want to explain “ There is no free lunch ” That's the truth ）
There are two things to understand when pushing the formula to the middle ：

1. sigma Symbols want to exchange order , The premise is that the variables are independent of each other
2. Because of the assumption that f It's evenly distributed , And it is a binary classification problem , All... No matter h There will be everything 1/2 The objective function value of quantity is equal to h Function values are equal .

Last , Found out “ No free lunch ” Theorem ——No Free Lunch Theorem.

4. Pumpkin book chapter 2

4.1 The formula 2.20 understand

Understand this formula , In fact, we want to know what is AUC, As long as we know ROC How did the curve get ,AUC It's natural to know ;
ROC The way the curve is drawn ： First, the predicted value of the sample is used as the keyword to sort from large to small ; Of the axis x The axis is the false positive case rate ,y The axis is the real case rate ; Draw the first dot with 1 Is the threshold , The second point takes the predicted value of the first sample as the threshold ; When you encounter positive sample points, point up a point , When you encounter negative sample points, draw a point to the right ( This is just experience ); Rigorous drawing needs to follow the formula .（ At the same time, a positive , A negative , Draw one on the top right ）

4.2 The formula 2.21 understand

Understanding method 1 ： It can be understood according to the pumpkin book
Understanding method 2 ： I find it understandable without simplification ; First of all, we still need to put 1/2 extracted ; Then it can be understood that the unit distance is 1, Next to the last bracket 1/m+ and 1/m- Is the correction of unit distance ;（ In essence, it is the trapezoidal area calculation method ）

4.3 The formula 2.27 understand ( await a vacancy or job opening )

Forget the confidence interval

4.4 The formula 2.41 understand ( await a vacancy or job opening )