特征值预处理

悠悠 2021-12-05 00:51 343阅读 0赞

## 一.归一化的概念 ##

### 　　　　1.概念 ###

特点：
        通过对原始数据的变换映射到默认为[0,1]之间
    目的：是的某一特征值不会对结果造成更大的影响===》几个特征值对结果影响权重相等的二十号要进行归一化
    
    缺点：异常点（在最大最小值之外）
    　　对异常点的处理不好，鲁棒性较差，只适合传统的小数据场景

### 　　2.使用　　 ###

![1513642-20190713212905003-1110263977.png][]

![1513642-20190713212923752-1134517683.png][]

实例：

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    #author tom
    #导入归一化处理的包
    from sklearn.preprocessing import MinMaxScaler
    
    def mm(x):
        """
        对数据预处理(归一化处理)
        :return:
        """
        min_max=MinMaxScaler()
        data=min_max.fit_transform(x)
        print(data)
    
    
    if __name__ == '__main__':
        l=[
            [90,2,10,40],
            [60,4,15,45],
            [75,3,13,46]]
        mm(l)

结果：

![1513642-20190713215814287-279222998.png][]

改变归一化范围：

![1513642-20190713215955474-733431370.png][]

## 二.标准化 ##

### 　　1.标准化的相关知识 ###

特点：方差越小数据越集中，方差越大越分散
            在样本足够多的时候稳定，适合现代嘈杂的大数据场景

![1513642-20190713223913940-1660880377.png][]

![1513642-20190713224006048-622630962.png][]

![1513642-20190713224035950-2093838425.png][]

![1513642-20190713224046697-160756373.png][]

实例：

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    #author tom
    
    from  sklearn.preprocessing import StandardScaler
    
    
    def stander(x):
        """
        标准化缩放
        :param x:
        :return:
        """
        s=StandardScaler()
        data=s.fit_transform(x)
        print (data)
    
    if __name__ == '__main__':
        l=[[1.,-1.,3.],
           [2.,4.,2.],
           [4.,6.,-1,]
           ]
        stander(l)

结果：

![1513642-20190713224116176-1429053363.png][]

## 3.缺失值处理　　 ##

![1513642-20190713225431076-1588552272.png][]

![1513642-20190713225447779-2111922590.png][]

![1513642-20190713225459027-550417945.png][]

![1513642-20190713225515574-586564461.png][]

实例：

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    #author tom
    from sklearn.preprocessing import Imputer
    import numpy as np
    
    def im(l):
        """
        缺失值的处理
        :return:
        """
        #NAN nan都可以 用平均值替换
        im=Imputer(missing_values='NaN',strategy='mean',axis=0)
        data=im.fit_transform(l)
        print(data)
    if __name__ == '__main__':
        l=[[1,3],
           [np.nan,2],
           [4,6]]
        im(l)

结果：

![1513642-20190713230552609-1248266639.png][]

注意：缺失值的形式一定是np.nan

![1513642-20190713230824814-1722958989.png][]

![1513642-20190713230834199-137808053.png][]

转载于:https://www.cnblogs.com/tjp40922/p/11182165.html

[1513642-20190713212905003-1110263977.png]: /images/20211205/a50f0682d8f34c0ba38893b904545db2.png
[1513642-20190713212923752-1134517683.png]: /images/20211205/635d045630134115a714a765492d57c0.png
[1513642-20190713215814287-279222998.png]: /images/20211205/4d05e4086bf64169981df14cf6b55562.png
[1513642-20190713215955474-733431370.png]: /images/20211205/5c0eda89401948b39bceacef3a3f3d56.png
[1513642-20190713223913940-1660880377.png]: /images/20211205/8ab96eabfdbe4d75a162ea92977e731e.png
[1513642-20190713224006048-622630962.png]: /images/20211205/e9c7ed553fa1458a972b682481a149e4.png
[1513642-20190713224035950-2093838425.png]: /images/20211205/41bf1052d40e41a7b1321c8e6d788cab.png
[1513642-20190713224046697-160756373.png]: /images/20211205/ba5fddf7596f4e42bbf43209a459bec6.png
[1513642-20190713224116176-1429053363.png]: /images/20211205/4029fe08bca241869a685520f1ab2f62.png
[1513642-20190713225431076-1588552272.png]: /images/20211205/485a41a4b39c4b32b4e03cfb3b61fa90.png
[1513642-20190713225447779-2111922590.png]: /images/20211205/e860f96eb7de4ff883126f368a1adfd8.png
[1513642-20190713225459027-550417945.png]: /images/20211205/af147ca2783145789c9b1084d6422499.png
[1513642-20190713225515574-586564461.png]: /images/20211205/7e2b29057acc47cf9b9302efef55287a.png
[1513642-20190713230552609-1248266639.png]: /images/20211205/df25135cb1024008bb9beffb00fa3020.png
[1513642-20190713230824814-1722958989.png]: /images/20211205/9bc75adba1b7406aa5830f4f34e121d9.png
[1513642-20190713230834199-137808053.png]: /images/20211205/bfa86a0e583649b78ad518eab688477d.png