【Python】数据预处理：归一化和离散化（MinMaxScaler,StandardScaler,cut,qcut）

秒速五厘米 2023-02-09 10:27 185阅读 0赞

## 一、内容来源 ##

课程来源：[大数据分析师（第一期）（学堂在线 北邮 杨亚）][Link 1]

数据集分享：[https://pan.baidu.com/s/197wLiuRLaB8kOxV9Ay7eTQ][https_pan.baidu.com_s_197wLiuRLaB8kOxV9Ay7eTQ]  
提取码：hldd

## 二、学习笔记 ##

![在这里插入图片描述][20200518175415214.png_x-oss-process_image]

![在这里插入图片描述][20200518175336641.png_x-oss-process_image]

**注意点：**  
若采用sklearn.preprocessing中的函数进行归一化，要将数据转为二维的！数据只有一维，可采用方法：reshape(-1, 1) \#变为n行1列的二维矩阵形式

参考资料：[25、pandas的reshape(1,-1)什么意思？][25_pandas_reshape_1_-1]

## 三、代码分享 ##

import pandas as pd
    
    filename = 'gz3.csv'
    df = pd.read_csv(filename,encoding='utf-8',usecols=[12,11])
    #filename = 'sy3.csv'
    #df = pd.read_csv(filename,encoding='utf-8',usecols=[22,23])
    
    print(df.head())
    print(df.describe())
    print(df.info())
    
    import matplotlib.pyplot as plt
    fig = plt.figure()
    ax1 = fig.add_subplot(121)
    ax2 = fig.add_subplot(122)
    x1 =  df['HUMI_new2']
    y1 =  df['PRES_new2']
    
    
    def count_elements(scores): #定义转换函数，统计每个数值对应多少个
        scorescount = {
        } #定义一个字典对象
        for i in scores :
            scorescount[int(i)] = scorescount.get(int(i), 0) + 1 #累加每个分数值的人数
        return scorescount
    
    counted1 = count_elements(x1)
    counted2 = count_elements(y1)
    
    ax1.set_title('HUMI')
    ax1.bar(counted1.keys(),counted1.values(),0.8,alpha=0.5,color='b')
    
    ax1.set_title('PRES')
    ax2.bar(counted2.keys(),counted2.values(),0.8,alpha=0.5,color='r')

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70]

import pandas as pd
    import matplotlib.pyplot as plt
    filename = 'gz3.csv'
    df = pd.read_csv(filename,encoding='utf-8',usecols=[12,11])
    
    fig = plt.figure()
    ax1 = fig.add_subplot(231)
    
    #子图1：原始图像
    x1 = df['PRES_new2']
    y1 = df['HUMI_new2']
    ax1.scatter(x1, y1, s=10)
    #ax1.set_xlim(0,1250)
    #ax1.set_ylim(0,120)
    ax1.set_title('Original')
    
    
    #子图2:(0,1)归一化，采用MinMaxScaler函数
    '''
    min=x1.min()
    max=x1.max( )
    ave=x1.mean()
    std =x1.std()
    x2=(x1-min)/(max-min)
    '''
    ax2 = fig.add_subplot(232)
    from sklearn.preprocessing import MinMaxScaler
    scaler = MinMaxScaler()
    x_reshape = x1.values.reshape(-1, 1) #变为n行1列的二维矩阵形式
    x2 = scaler.fit_transform(x_reshape) #调用MinMaxScaler的fit.transform转换方法，进行归一化处理
    y_reshape = y1.values.reshape(-1, 1) #变为n行1列的二维矩阵形式
    y2 = scaler.fit_transform(y_reshape) #调用MinMaxScaler的fit.transform转换方法，进行归一化处理
    ax2.scatter(x2, y2, s=10)
    ax2.set_title('MinMaxScaler')
    
    
    
    #子图3：Z-score归一化  采用StandardScaler()函数
    ax3 = fig.add_subplot(233)
    from sklearn.preprocessing import StandardScaler
    scaler = StandardScaler()
    x_reshape = x1.values.reshape(-1, 1) #变为n行1列的二维矩阵形式
    x3 = scaler.fit_transform(x_reshape) #调用MinMaxScaler的fit.transform转换方法，进行归一化处理
    y_reshape = y1.values.reshape(-1, 1) #变为n行1列的二维矩阵形式
    y3 = scaler.fit_transform(y_reshape) #调用MinMaxScaler的fit.transform转换方法，进行归一化处理
    ax3.scatter(x3, y3, s=10)
    ax3.set_title('StandardScaler')
    
    
    def count_elements(scores): #定义转换函数，统计每个数值对应多少个
        scorescount = {
        } #定义一个字典对象
        for i in scores : #累加每个分数值的人数
            scorescount[int(i)] = scorescount.get(int(i), 0) + 1 #get(返回字典中的值，默认值)
        return scorescount
    
    def count_elements1(scores): #扩大计数范围，使得效果更明显
        scorescount = {
        } #定义一个字典对象
        for i in scores : #累加每个分数值的人数
            scorescount[int(i*100)] = scorescount.get(int(i*100), 0) + 1 #get(返回字典中的值，默认值)
        return scorescount
    
    ax4 = fig.add_subplot(234)
    ax5 = fig.add_subplot(235)
    ax6 = fig.add_subplot(236)
    ax4.set_title('ori')
    ax5.set_title('minmax')
    ax6.set_title('standard')
    counted1 = count_elements(x1)
    counted2 = count_elements1(x2)
    counted3 = count_elements1(x3)
    ax4.bar(counted1.keys(),counted1.values(),0.8,alpha=0.5,color='r')
    ax5.bar(counted2.keys(),counted2.values(),0.8,alpha=0.5,color='r')
    ax6.bar(counted3.keys(),counted3.values(),0.8,alpha=0.5,color='r')

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 1]

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 2]

--------------------

## 补充·离散化 ##

简略记录：  
![在这里插入图片描述][20200521004518379.png_x-oss-process_image]

![在这里插入图片描述][20200521004809354.png_x-oss-process_image]

#连续值的离散化
    import numpy as np
    import pandas as pd
    data=pd.Series(np.arange(11))
    result1=pd.cut(data,5)
    result2=pd.cut(data,4)
    bins=[0,2,7,10]
    result3=pd.cut(data,bins)
    print('--------------result1-----------------')
    print(result1)
    print('--------------result1 count-----------------')
    print(pd.value_counts(result1))
    
    print('--------------result2-----------------')
    print(result2)
    print('--------------result2 count-----------------')
    print(pd.value_counts(result2))
    
    print('--------------result3-----------------')
    print(result3)
    print('--------------result3 count-----------------')
    print(pd.value_counts(result3))
    
    data=pd.Series([0,11,18,23,40,50,69,73,88])
    result1=pd.cut(data,4)
    result2=pd.qcut(data,4)
    
    print('--------------result1-----------------')
    print(result1)
    print('--------------result1 count-----------------')
    print(pd.value_counts(result1))
    
    print('--------------result2-----------------')
    print(result2)
    print('--------------result2 count-----------------')
    print(pd.value_counts(result2))
    
    
    file="sy3.csv"
    df=pd.read_csv(file,encoding='utf-8',usecols=[20])
    print('--------------head-----------------')
    print(df.head())
    print('--------------describe-----------------')
    print(df.describe())
    print('--------------info-----------------')
    print(df.info)
    
    sections=[0,50,100,150,200,300,1200]#设置划分区间
    sections_name=['green','yellow','orange','red','purple','Brownish red']
    result=pd.cut(df.ave,sections,labels=sections_name)
    
    print('--------------result-----------------')
    print(result)
    print('--------------result type-----------------')
    print(type(result))
    print('--------------result count-----------------')
    print(pd.value_counts(result))
    
    result1=pd.cut(df.ave,4)
    print('--------------result1-----------------')
    print(result1.head())
    print('--------------result count-----------------')
    print(pd.value_counts(result1))

部分结果：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 3]

[Link 1]: http://www.xuetangx.com/courses/course-v1:xuetangx+BigDataBA01+2019_T2/about
[https_pan.baidu.com_s_197wLiuRLaB8kOxV9Ay7eTQ]: https://pan.baidu.com/s/197wLiuRLaB8kOxV9Ay7eTQ
[20200518175415214.png_x-oss-process_image]: https://img-blog.csdnimg.cn/20200518175415214.png?x-oss-process=image
[20200518175336641.png_x-oss-process_image]: https://img-blog.csdnimg.cn/20200518175336641.png?x-oss-process=image
[25_pandas_reshape_1_-1]: https://www.jianshu.com/p/d9df005636a6
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20200518153324572.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20200518174948886.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20200518175028278.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz,size_16,color_FFFFFF,t_70
[20200521004518379.png_x-oss-process_image]: https://img-blog.csdnimg.cn/20200521004518379.png?x-oss-process=image
[20200521004809354.png_x-oss-process_image]: https://img-blog.csdnimg.cn/20200521004809354.png?x-oss-process=image
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20200521004931271.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz,size_16,color_FFFFFF,t_70