Chapter one Feature Engineering

As the name suggests, feature engineering is the engineering processing of raw data , Refine it into characteristics , As input for algorithm and model implementation .
This chapter mainly discusses two common data types ：

1. Structured data . Structured data type can be regarded as a table of relational database , There is a clear definition , It includes numerical type , There are two basic types of category type , Each row of data represents the information of a sample
2. Unstructured data . Unstructured sleep mainly includes text , Images , Sound evaluation , Video data , It contains information that cannot be represented by a simple number , There is no clear definition of categories , And the size of each data is different

01 Feature normalization

In order to eliminate the dimensional impact between data features , We need to normalize the features , Make different indicators comparable . for example , Analyze the impact of a person's height and weight on health , If you use rice m And kilogram kg Work unit , Then the height characteristics will be in 1.6-1.8m Within the numerical range of , Weight will be 50-100kg Within the scope of , The results of the analysis will tend to the weight characteristics with large numerical differences . Want to get more accurate demerit recording , We need to normalize the features （Normalization） Handle , Make each index in the same numerical magnitude , For analysis

1. Why do we need to normalize the characteristics of numerical types

Normalization of features such as logarithm can mean that all features are unified into a roughly the same numerical range .
There are two common methods

1. Normalization of linear function Min-Max Scaling
   It is more linear to the original data than that , Map the results to 【0,1】 The scope of the , Realize the scaling of the original data

   （x-xmin）/ （xmax-xmin）

2. Zero mean normalization Z-Score Normalization
   It will say that the original data is mapped to an average of 0, The standard deviation is 1 The distribution of , say concretely , Assume that the mean value of the original feature is u, The standard deviation is m ：

   （x-u） / m

Why do we need to normalize numerical features ？ It is advisable to illustrate the importance of normalization with the help of practical examples of random gradient descent , Suppose there are two numerical characteristics ,x1 Range greater than x2.
At the same learning rate ,x1 The update speed of will be greater than x2, It takes more iterations to find the optimal solution , If you will x1 and x2 After normalization to the same numerical range , The contour map of the optimization target will change from ellipse to circle ,x1 and x2 The update speed of becomes more consistent , It is easy and faster to find the optimal solution through gradient descent

Of course , Data normalization is not everything , in application , The model solved by gradient descent method usually needs normalization , Including linear regression , Logical regression , Vector machine only , Neural networks and other models , But it is not suitable for decision tree model , The node classification of decision tree is mainly based on the data set D About the characteristics of x The information gain ratio of , And the information gain ratio has nothing to do with whether the feature is normalized , Because normalization does not change the characteristics of the sample x Information gain on