Yu-Dem~~

php正则匹配sg-nc-wap_Pandas基础操作（上）

文章目录

一、Pandas文件读取

1.pandas数据读取

1、读取纯文本文件

1.1 读取csv，使用默认的标题行、逗号分隔符

1.2 读取txt文件，自己指定分隔符、列名

2、读取excel文件

3、读取sql文件

二、pandas的数据结构DataFrame和Series

DataFrame：二维数据，整个表格，多行多列

1.Series

1.1 仅有数据列表即可生产最简单的Series

1.2 创建一个具有标签索引的Series

1.3 使用python字典创建Series

1.4 根据数据标签索引查询数据

2. DataFrame

2.1 根据多个字典序列创建DataFrame

从DataFrame中查询出Series

3.1 查询一列,结果是一个pd.Series

3.2 查询多列,结果是一个pd.DataFrame

3.3 查询一行,结果是一个pd.Series

3.4 查询多行,结果是一个pd.DataFrame

三.Pandas查询数据的5种方法

Pandas查询数据的几种方法

Pandas使用df.loc查询数据的方法

注意

0. 读取数据

1. 使用单个label值查询数据

2. 使用值列进行表批量查询

3. 使用数值区间进行范围查询

4. 使用条件表达式查询

复杂条件查询，查询一下完美得天气

5. 调用函数查询

四、Pandas怎样新增数据列

0. 读取csv数据到DataFrame

1. 直接赋值的方法

2. df.apply方法

3. df.assign方法

4. 按条件选择分组分别赋值

五、Pandas数据统计函数

0. 读取csv数据

1. 汇总类统计

2. 唯一去重和按值计数

2.1 唯一去重

2.2 按值计数

3. 相关系数和协防差

六、Pandas对缺失值的处理

实例：特殊excel的读取、清洗、处理

步骤1：读取excel的时候，忽略前几个空行

步骤2：检测空值

步骤3：删除掉全是空值的列

步骤4：删除掉全是空值的行

步骤5：将分数列为空的填充为0分

步骤6：将姓名的缺失值填充

步骤7：将清晰好的excel保存

七、Pandas的SettingWithCopyWarning报警

0. 数据读取

1. 复现

2、原因

4. 解决方法2

Pandas不允许先筛选子DataFrame，在进行修改写入

八、Pandas数据排序

0. 读取数据

1. Series的排序

2. DataFrame的排序

2.1 单列排序

2.2 多列排序

九、Pandas字符串处理

0. 读取北京2018年天气数据

1. 获取Series的str属性，使用各种字符串处理函数

4. 使用正则表达式的处理

Series.str默认就开启了正则表达式模式

十、Pandas的axis参数怎么理解？

***按哪个axis，就是这个axis要动起来(类似被for遍历)，其它的axis保持不动\***

1. 单列drop, 就是删除某一列

3. 按照axis=0/index执行mean聚合操作

***按哪个axis，就是这个axis要动起来(类似被for遍历)，其它的axis保持不动\***

3. 按照axis=1/columns执行mean聚合操作

***按哪个axis，就是这个axis要动起来(类似被for遍历)，其它的axis保持不动\***

5. 再次举例, 加深理解

***按哪个axis，就是这个axis要动起来(类似被for遍历)，其它的axis保持不动\***

十一、Pandas的索引index的用途

1. 使用index查询数据

2. 使用index会提升查询性能

实验1:完全随机的查询顺序

实验2:将index排序后的查询

3.使用index能自动对齐数据

s1,s2都具有b,c索引,而a,d为各自独有,无法对齐,所有相加结果为空

4. 使用index更多更强大的数据结构支持

十二、Pandas怎样实现DataFrame的Merge

merge的语法：

1、电影数据集的join实例

电影评分数据集

2、理解merge时数量的对齐关系

2.1 one-to-one 一对一关系的merge

2.2 one-to-many 一对多关系的merge

2.3 many-to-many 多对多关系的merge

3、理解left join、right join、inner join、outer join的区别

3.1 inner join，默认

3.2 left join

3.3 right join

3.4 outer join

4、如果出现非Key的字段重名怎么办

十三、Pandas实现数据的合并concat

使用场景：

一句话说明concat语法：

concat语法：pandas.concat(objs, axis=0, join='outer', ignore_index=False)

append语法：DataFrame.append(other, ignore_index=False)

参考文档：

一、使用Pandas.concat合并数据

1. 默认的concat, 参数为axis=0, join=outer, ignore_index=False

2. 使用ignore_index=True可以忽略原来的索引

3. 使用join=inner过滤掉不匹配的列

4. 使用axis=1相当于添加新列

A:添加一列Series

B:添加多列Series

二、使用DateFrame.append按行合并数据

1. 给一个DataFrame添加另一个DataFrame

2. 忽略原来的索引，另ignore_index=True

3.可以一行一行的给DataFrame添加数据

A:低性能版

B:性能好的版本

十四、Pandas批量拆分Excel与合并Excel

0. 读取源Excel到Pandas

1、将一个大excel等份拆成多个Excel

1. 1 计算拆分后的每个excel的行数

1.2 拆分成多个DataFrame

1.3 将每个DataFrame存入excel

2、合并多个小Excel到一个大Excel

2.1 遍历文件夹，得到要合并的Excel名称列表

2.2 分别读取到DataFrame

2.3 使用pd.concat进行合并

2.4 将合并后的DataFrame输出到Excel

十五、Pandas怎样实现groupby分组统计

1、分组使用聚合函数做数据统计

1.1 单个列groupby，查询所有数据列的统计

1.2 多个列groupby，查询所有数据列的统计

1.3 同时查看多种数据统计

1.4 查看单列的结果数据统计

1.5 不同列使用不同的聚合函数

2、遍历groupby的结果理解执行流程

2.1 遍历单个列聚合的分组

2.2 遍历多个列聚合的分组

3、实例分组探索天气数据

3.1 查看每个月的最高温度

3.2 查看每个月的最高温度、最低温度、平均空气质量指数

十六、Pandas的分层索引MultiIndex

1、Series的分层索引MultiIndex

2、Series有多层索引MultiIndex怎么筛选数据？

3、DataFrame的多层索引MultiIndex

4、DataFrame有多层索引MultiIndex怎样筛选？

十七、Pandas的数据转换函数map、apply、applymap

1. map用于Series值的转换

方法1：Series.map(dict)

方法2：Series.map(function)

2. apply用于Series和DataFrame的转换

Series.apply(function)

DataFrame.apply(function)

3. applymap用于DataFrame所有值的转换

十八、Pandas怎样对每个分组应用apply函数?

知识：Pandas的GroupBy遵从split、apply、combine模式

GroupBy.apply(function)

本次实例演示：

实例1：怎样对数值列按分组的归一化？

演示：用户对电影评分的归一化

实例2：怎么取每个分组的TOP N数据

一、Pandas文件读取

1.pandas数据读取

pandas需要先读取表格类型的数据，然后进行分析

数据类型

说明

pandas读取方法

csv、tsv、txt

用逗号分隔、tab分割的纯文本文件

pd.read_csv

excel

微软xls或者xlsx文件

pd.read_excel

mysql

关系型数据库表

pd.read_sql

In [1]:

import pandas as pd

1、读取纯文本文件

1.1 读取csv，使用默认的标题行、逗号分隔符

In [2]:

fpath = "./pandas-learn-code/datas/ml-latest-small/ratings.csv"

In [3]:

# 使用pd.read_csv读取数据

ratings = pd.read_csv(fpath)

In [4]:

# 查看前几行数据

ratings.head()

Out[4]:

userId

movieId

rating

timestamp

4.0

964982703

4.0

964981247

4.0

964982224

5.0

964983815

5.0

964982931

In [5]:

# 查看数据的形状，返回(行数、列数)

ratings.shape

Out[5]:

(100836, 4)

In [6]:

# 查看列名列表

ratings.columns

Out[6]:

Index(['userId', 'movieId', 'rating', 'timestamp'], dtype='object')

In [7]:

# 查看索引

ratings.index

Out[7]:

RangeIndex(start=0, stop=100836, step=1)

In [9]:

# 查看每列的数据类型

ratings.dtypes

Out[9]:

userId int64

movieId int64

rating float64

timestamp int64

dtype: object

1.2 读取txt文件，自己指定分隔符、列名

In [10]:

fpath = "./pandas-learn-code/datas/crazyant/access_pvuv.txt"

In [11]:

pvuv = pd.read_csv(fpath, sep="\t", header=None, names=["pdate","pv","uv"])

sep代表分隔符

header=none代表没有列名

names代表指定的列明

In [13]:

pvuv.head()

Out[13]:

pdate

2019-09-10

139

2019-09-09

185

153

2019-09-08

123

2019-09-07

2019-09-06

157

2、读取excel文件

In [18]:

fpath = "./pandas-learn-code/datas/crazyant/access_pvuv.xlsx"

pvuv = pd.read_excel(fpath)

In [19]:

pvuv

Out[19]:

日期

2019-09-10

139

2019-09-09

185

153

2019-09-08

123

2019-09-07

2019-09-06

157

2019-09-05

205

151

2019-09-04

196

167

2019-09-03

216

176

2019-09-02

227

148

2019-09-01

105

3、读取sql文件

In [36]:

import pymysql

conn = pymysql.connect(

host="127.0.0.1",

user="root",

password="123456",

database="test",

charset="utf8"

)

In [41]:

fpath = "./pandas-learn-code/datas/crazyant/test_crazyant_pvuv.sql"

mysql_page = pd.read_sql("select * from crazyant_pvuv", con=conn)

In [42]:

pvuv

Out[42]:

日期

2019-09-10

139

2019-09-09

185

153

2019-09-08

123

2019-09-07

2019-09-06

157

2019-09-05

205

151

2019-09-04

196

167

2019-09-03

216

176

2019-09-02

227

148

2019-09-01

105

二、pandas的数据结构DataFrame和Series

DataFrame：二维数据，整个表格，多行多列

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-tT4RRssV-1597761927694)(C:\Users\z&y\AppData\Roaming\Typora\typora-user-images\image-20200730213558995.png)]

In [1]:

import pandas as pd

import numpy as np

1.Series

Series是一种类似于一维数组的对象，它由一组数据(不同数据类型)以及一组与之相关的数据标签(即索引)组成。

1.1 仅有数据列表即可生产最简单的Series

In [3]:

s1 = pd.Series([1,'a',5.2,7])

In [5]:

# 左侧为索引，右侧是数据

s1.head()

Out[5]:

0 1

1 a

2 5.2

3 7

dtype: object

In [6]:

# 获取索引

s1.index

Out[6]:

RangeIndex(start=0, stop=4, step=1)

In [7]:

# 获取数据

s1.values

Out[7]:

array([1, 'a', 5.2, 7], dtype=object)

1.2 创建一个具有标签索引的Series

In [8]:

s2 = pd.Series([1,'a',5.2,7], index=['a','b','c','d'])

In [9]:

Out[9]:

a 1

b a

c 5.2

d 7

dtype: object

In [10]:

s2.index

Out[10]:

Index(['a', 'b', 'c', 'd'], dtype='object')

1.3 使用python字典创建Series

In [11]:

sdata = {'Ohio':35000, 'Texas':72000, 'Oregon':16000, 'Utah':5000}

In [13]:

s3 = pd.Series(sdata)

In [14]:

# 字典的key成为了Series的索引

Out[14]:

Ohio 35000

Texas 72000

Oregon 16000

Utah 5000

dtype: int64

1.4 根据数据标签索引查询数据

类似python的字典dict

In [15]:

Out[15]:

a 1

b a

c 5.2

d 7

dtype: object

In [20]:

s2['a']

Out[20]:

In [21]:

# 查询一个值,返回查询值的数据类型

type(s2['a'])

Out[21]:

int

In [18]:

# 一次查询多个值

s2[['a','b','c']]

Out[18]:

a 1

b a

c 5.2

dtype: object

In [22]:

# 查询多个值,返回的还是Series

type(s2[['a','b','c']])

Out[22]:

pandas.core.series.Series

2. DataFrame

DataFrame是一个表格型的数据结构

每列可以是不同的值类型(数值,字符串,布尔值等)

既有行索引index,也有列索引columns

可以被看做由Series组成的字典

2.1 根据多个字典序列创建DataFrame

In [24]:

data = {

'state':['Ohio','Ohio','Ohio','Nevada','Nevada'],

'year':[2000,2001,2002,2003,2004],

'pop':[1.5,1.7,3.6,2.4,2.9]

}

df = pd.DataFrame(data)

In [25]:

Out[25]:

state

year

pop

Ohio

2000

1.5

Ohio

2001

1.7

Ohio

2002

3.6

Nevada

2003

2.4

Nevada

2004

2.9

In [26]:

df.dtypes

Out[26]:

state object

year int64

pop float64

dtype: object

In [27]:

df.columns

Out[27]:

Index(['state', 'year', 'pop'], dtype='object')

In [28]:

df.index

Out[28]:

RangeIndex(start=0, stop=5, step=1)

从DataFrame中查询出Series

如果只查询一列,一列,返回的是pd.Series

如果查询多行,多列,返回的是pd.DataFrame

In [29]:

Out[29]:

state

year

pop

Ohio

2000

1.5

Ohio

2001

1.7

Ohio

2002

3.6

Nevada

2003

2.4

Nevada

2004

2.9

3.1 查询一列,结果是一个pd.Series

In [30]:

df['year']

Out[30]:

0 2000

1 2001

2 2002

3 2003

4 2004

Name: year, dtype: int64

In [35]:

# 返回的是一个Series

type(df['year'])

Out[35]:

pandas.core.series.Series

3.2 查询多列,结果是一个pd.DataFrame

In [33]:

df[['year', 'pop']]

Out[33]:

year

pop

2000

1.5

2001

1.7

2002

3.6

2003

2.4

2004

2.9

In [34]:

# 返回的结果是一个DataFrame

type(df[['year','pop']])

Out[34]:

pandas.core.frame.DataFrame

3.3 查询一行,结果是一个pd.Series

In [39]:

df.loc[0]

Out[39]:

state Ohio

year 2000

pop 1.5

Name: 0, dtype: object

In [40]:

type(df.loc[0])

Out[40]:

pandas.core.series.Series

3.4 查询多行,结果是一个pd.DataFrame

In [41]:

# DataFrame中切片会返回结尾的数据

df.loc[0:3]

Out[41]:

state

year

pop

Ohio

2000

1.5

Ohio

2001

1.7

Ohio

2002

3.6

Nevada

2003

2.4

In [42]:

type(df.loc[0:3])

Out[42]:

pandas.core.frame.DataFrame

三.Pandas查询数据的5种方法

Pandas查询数据的几种方法

df.loc方法,根据行,列的标签值查询

df.iloc方法,根据行,列的数字位置查询

df.where方法

df.query方法

.loc方法既能查询,又能覆盖写入,推荐使用此方法

Pandas使用df.loc查询数据的方法

使用单个label值查询数据

使用值列表批量查询

使用数值区间进行范围查询

使用条件表达式查询

调用函数查询

注意

以上查询方法,既适用于行,也适用于列

In [3]:

import pandas as pd

0. 读取数据

数据为北京2018年全年天气预报

In [4]:

df = pd.read_csv("./pandas-learn-code/datas/beijing_tianqi/beijing_tianqi_2018.csv")

In [5]:

df.head()

Out[5]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

2018-01-01

3℃

-6℃

晴~多云

东北风

1-2级

良

2018-01-02

2℃

-5℃

阴~多云

东北风

1-2级

优

2018-01-03

2℃

-5℃

多云

北风

1-2级

优

2018-01-04

0℃

-8℃

阴

东北风

1-2级

优

2018-01-05

3℃

-6℃

多云~晴

西北风

1-2级

优

In [6]:

# 设定索引为日期,方便按日期筛选

df.set_index('ymd', inplace=True)

In [7]:

df.head()

Out[7]:

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

ymd

2018-01-01

3℃

-6℃

晴~多云

东北风

1-2级

良

2018-01-02

2℃

-5℃

阴~多云

东北风

1-2级

优

2018-01-03

2℃

-5℃

多云

北风

1-2级

优

2018-01-04

0℃

-8℃

阴

东北风

1-2级

优

2018-01-05

3℃

-6℃

多云~晴

西北风

1-2级

优

In [8]:

# 时间序列见后续课程,本次按字符串处理

df.index

Out[8]:

Index(['2018-01-01', '2018-01-02', '2018-01-03', '2018-01-04', '2018-01-05',

'2018-01-06', '2018-01-07', '2018-01-08', '2018-01-09', '2018-01-10',

...

'2018-12-22', '2018-12-23', '2018-12-24', '2018-12-25', '2018-12-26',

'2018-12-27', '2018-12-28', '2018-12-29', '2018-12-30', '2018-12-31'],

dtype='object', name='ymd', length=365)

In [9]:

# 替换掉温度的后缀℃

# df.loc[:]表示筛选出所有的行

df.loc[:, "bWendu"] = df["bWendu"].str.replace("℃","").astype('int32')

df.loc[:, "yWendu"] = df["yWendu"].str.replace("℃","").astype('int32')

In [10]:

# bWendu和yWendu改为int类型

df.dtypes

Out[10]:

bWendu int32

yWendu int32

tianqi object

fengxiang object

fengli object

aqi int64

aqiInfo object

aqiLevel int64

dtype: object

1. 使用单个label值查询数据

行或者列,都可以只传入单个值,实现精确匹配

In [11]:

# 得到单个值

df.loc['2018-01-03','bWendu']

Out[11]:

In [12]:

# 得到一个Series

df.loc['2018-01-03',['bWendu', 'yWendu']]

Out[12]:

bWendu 2

yWendu -5

Name: 2018-01-03, dtype: object

2. 使用值列进行表批量查询

In [13]:

# 得到Series

df.loc[['2018-01-03','2018-01-04','2018-01-05'], 'bWendu']

Out[13]:

ymd

2018-01-03 2

2018-01-04 0

2018-01-05 3

Name: bWendu, dtype: int32

In [14]:

# 得到DataFrame

df.loc[['2018-01-03','2018-01-04','2018-01-05'], ['bWendu','yWendu']]

Out[14]:

bWendu

yWendu

ymd

2018-01-03

-5

2018-01-04

-8

2018-01-05

-6

3. 使用数值区间进行范围查询

注意:区间既包含开始,也包含结束

In [15]:

# 行index按区间

df.loc['2018-01-03':'2018-01-05', 'bWendu']

Out[15]:

ymd

2018-01-03 2

2018-01-04 0

2018-01-05 3

Name: bWendu, dtype: int32

In [16]:

# 列index按区间

df.loc['2018-01-03','bWendu':'fengxiang']

Out[16]:

bWendu 2

yWendu -5

tianqi 多云

fengxiang 北风

Name: 2018-01-03, dtype: object

In [17]:

# 行和列都按区间查询

df.loc['2018-01-03':'2018-01-05','bWendu':'fengxiang']

Out[17]:

bWendu

yWendu

tianqi

fengxiang

ymd

2018-01-03

-5

多云

北风

2018-01-04

-8

阴

东北风

2018-01-05

-6

多云~晴

西北风

4. 使用条件表达式查询

bool列表的长度得等于行数或者列数

In [23]:

df.loc[df["yWendu"]

Out[23]:

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

ymd

2018-01-23

-4

-12

晴

西北风

3-4级

优

2018-01-24

-4

-11

晴

西南风

1-2级

优

2018-01-25

-3

-11

多云

东北风

1-2级

优

2018-12-26

-2

-11

晴~多云

东北风

2级

优

2018-12-27

-5

-12

多云~晴

西北风

3级

优

2018-12-28

-3

-11

晴

西北风

3级

优

2018-12-29

-3

-12

晴

西北风

2级

优

2018-12-30

-2

-11

晴~多云

东北风

1级

优

In [24]:

df["yWendu"]

Out[24]:

ymd

2018-01-01 False

2018-01-02 False

2018-01-03 False

2018-01-04 False

2018-01-05 False

2018-01-06 False

2018-01-07 False

2018-01-08 False

2018-01-09 False

2018-01-10 False

2018-01-11 False

2018-01-12 False

2018-01-13 False

2018-01-14 False

2018-01-15 False

2018-01-16 False

2018-01-17 False

2018-01-18 False

2018-01-19 False

2018-01-20 False

2018-01-21 False

2018-01-22 False

2018-01-23 True

2018-01-24 True

2018-01-25 True

2018-01-26 False

2018-01-27 False

2018-01-28 False

2018-01-29 False

2018-01-30 False

...

2018-12-02 False

2018-12-03 False

2018-12-04 False

2018-12-05 False

2018-12-06 False

2018-12-07 False

2018-12-08 False

2018-12-09 False

2018-12-10 False

2018-12-11 False

2018-12-12 False

2018-12-13 False

2018-12-14 False

2018-12-15 False

2018-12-16 False

2018-12-17 False

2018-12-18 False

2018-12-19 False

2018-12-20 False

2018-12-21 False

2018-12-22 False

2018-12-23 False

2018-12-24 False

2018-12-25 False

2018-12-26 True

2018-12-27 True

2018-12-28 True

2018-12-29 True

2018-12-30 True

2018-12-31 False

Name: yWendu, Length: 365, dtype: bool

复杂条件查询，查询一下完美得天气

注意，组合条件用&符号合并，每个条件判断都得带括号

In [29]:

df.loc[(df["bWendu"]<=30) & (df["yWendu"]>=15) & (df["tianqi"]=="晴") & (df["aqiLevel"]==1),:]

Out[29]:

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

ymd

2018-08-24

晴

北风

1-2级

优

2018-09-07

晴

西北风

3-4级

优

In [30]:

(df["bWendu"]<=30) & (df["yWendu"]>=15) & (df["tianqi"]=="晴") & (df["aqiLevel"]==1)

Out[30]:

ymd

2018-01-01 False

2018-01-02 False

2018-01-03 False

2018-01-04 False

2018-01-05 False

2018-01-06 False

2018-01-07 False

2018-01-08 False

2018-01-09 False

2018-01-10 False

2018-01-11 False

2018-01-12 False

2018-01-13 False

2018-01-14 False

2018-01-15 False

2018-01-16 False

2018-01-17 False

2018-01-18 False

2018-01-19 False

2018-01-20 False

2018-01-21 False

2018-01-22 False

2018-01-23 False

2018-01-24 False

2018-01-25 False

2018-01-26 False

2018-01-27 False

2018-01-28 False

2018-01-29 False

2018-01-30 False

...

2018-12-02 False

2018-12-03 False

2018-12-04 False

2018-12-05 False

2018-12-06 False

2018-12-07 False

2018-12-08 False

2018-12-09 False

2018-12-10 False

2018-12-11 False

2018-12-12 False

2018-12-13 False

2018-12-14 False

2018-12-15 False

2018-12-16 False

2018-12-17 False

2018-12-18 False

2018-12-19 False

2018-12-20 False

2018-12-21 False

2018-12-22 False

2018-12-23 False

2018-12-24 False

2018-12-25 False

2018-12-26 False

2018-12-27 False

2018-12-28 False

2018-12-29 False

2018-12-30 False

2018-12-31 False

Length: 365, dtype: bool

5. 调用函数查询

In [31]:

# 直接写lambda表达式

df.loc[lambda df: (df["bWendu"]<=30) & (df["yWendu"]>=15),:]

Out[31]:

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

ymd

2018-04-28

晴

西南风

3-4级

125

轻度污染

2018-04-29

多云

南风

3-4级

193

中度污染

2018-05-04

晴~多云

西南风

1-2级

良

2018-05-09

晴~多云

西南风

3-4级

良

2018-05-10

多云

南风

3-4级

118

轻度污染

2018-05-11

阴~多云

东风

1-2级

106

轻度污染

2018-05-12

小雨

东南风

3-4级

186

中度污染

2018-05-13

晴

南风

1-2级

良

2018-05-16

多云~小雨

东风

1-2级

142

轻度污染

2018-05-17

小雨~多云

北风

1-2级

良

2018-05-18

多云~晴

南风

1-2级

优

2018-05-19

多云~小雨

南风

1-2级

良

2018-05-20

阴~小雨

东风

1-2级

良

2018-05-23

晴

西南风

3-4级

153

中度污染

2018-05-26

小雨~多云

西南风

3-4级

143

轻度污染

2018-05-28

晴

西北风

4-5级

178

中度污染

2018-06-09

小雨

北风

1-2级

优

2018-06-10

多云

东南风

1-2级

良

2018-06-11

多云

西南风

3-4级

良

2018-06-13

雷阵雨~多云

东北风

1-2级

良

2018-06-18

雷阵雨

西南风

1-2级

112

轻度污染

2018-06-22

雷阵雨~多云

东南风

1-2级

良

2018-07-08

雷阵雨

南风

1-2级

良

2018-07-09

雷阵雨~多云

东南风

1-2级

106

轻度污染

2018-07-10

多云~雷阵雨

南风

1-2级

优

2018-07-11

雷阵雨~大雨

东北风

1-2级

优

2018-07-12

多云

南风

1-2级

优

2018-07-13

雷阵雨

东风

1-2级

良

2018-07-17

中雨~雷阵雨

西风

1-2级

优

2018-07-24

暴雨~雷阵雨

东北风

3-4级

优

…

2018-08-11

雷阵雨~中雨

东风

1-2级

良

2018-08-12

雷阵雨

东南风

1-2级

良

2018-08-14

中雨~小雨

东北风

1-2级

优

2018-08-16

晴~多云

东北风

1-2级

优

2018-08-17

多云~雷阵雨

东南风

1-2级

良

2018-08-18

小雨~中雨

北风

3-4级

优

2018-08-19

中雨~小雨

东北风

1-2级

优

2018-08-22

雷阵雨~多云

西南风

1-2级

优

2018-08-24

晴

北风

1-2级

优

2018-08-27

多云~雷阵雨

东南风

1-2级

良

2018-08-28

小雨~多云

南风

1-2级

良

2018-08-30

多云

南风

1-2级

优

2018-08-31

多云~阴

东南风

1-2级

优

2018-09-01

阴~小雨

南风

1-2级

优

2018-09-02

小雨~多云

南风

1-2级

良

2018-09-03

晴

北风

3-4级

良

2018-09-06

多云~晴

西北风

4-5级

优

2018-09-07

晴

西北风

3-4级

优

2018-09-08

多云~晴

北风

1-2级

优

2018-09-09

晴

西南风

1-2级

良

2018-09-10

多云

南风

1-2级

良

2018-09-11

多云

南风

1-2级

良

2018-09-12

多云

南风

1-2级

良

2018-09-13

多云~阴

南风

1-2级

107

轻度污染

2018-09-14

小雨~多云

南风

1-2级

128

轻度污染

2018-09-15

多云

北风

3-4级

优

2018-09-17

多云~阴

北风

1-2级

优

2018-09-18

阴~多云

西南风

1-2级

优

2018-09-19

多云

南风

1-2级

良

2018-09-20

多云

西南风

1-2级

良

64 rows × 8 columns

In [33]:

# 编写自己的函数，查询9月份，空气质量好的数据

def query_my_data(df):

return df.index.str.startswith("2018-09") & (df["aqiLevel"]==1)

df.loc[query_my_data,:]

Out[33]:

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

ymd

2018-09-01

阴~小雨

南风

1-2级

优

2018-09-04

晴

西南风

3-4级

优

2018-09-05

晴~多云

西南风

3-4级

优

2018-09-06

多云~晴

西北风

4-5级

优

2018-09-07

晴

西北风

3-4级

优

2018-09-08

多云~晴

北风

1-2级

优

2018-09-15

多云

北风

3-4级

优

2018-09-16

多云~晴

北风

1-2级

优

2018-09-17

多云~阴

北风

1-2级

优

2018-09-18

阴~多云

西南风

1-2级

优

2018-09-21

晴

西北风

3-4级

优

2018-09-22

晴

西北风

3-4级

优

2018-09-23

晴

西北风

4-5级

优

2018-09-24

晴

北风

1-2级

优

2018-09-25

晴~多云

南风

1-2级

优

2018-09-29

晴

北风

3-4级

优

2018-09-30

多云

西北风

4-5级

优

四、Pandas怎样新增数据列

In [1]:

import pandas as pd

0. 读取csv数据到DataFrame

In [15]:

df = pd.read_csv("./pandas-learn-code/datas/beijing_tianqi/beijing_tianqi_2018.csv")

In [16]:

df.head()

Out[16]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

2018-01-01

3℃

-6℃

晴~多云

东北风

1-2级

良

2018-01-02

2℃

-5℃

阴~多云

东北风

1-2级

优

2018-01-03

2℃

-5℃

多云

北风

1-2级

优

2018-01-04

0℃

-8℃

阴

东北风

1-2级

优

2018-01-05

3℃

-6℃

多云~晴

西北风

1-2级

优

1. 直接赋值的方法

实例：清理温度列，变成数字类型

In [31]:

df.loc[:,"bWendu"] = df["bWendu"].str.replace("℃","").astype('int32')

df.loc[:,"yWendu"] = df["yWendu"].str.replace("℃","").astype('int32')

实例：计算温差

In [49]:

del df["bWendnu"]

In [51]:

del df["bWednu"]

In [52]:

# 注意，fpath["bWendu"]其实是一个Series，后面的减法返回的是Series

df.loc[:,"wencha"] = df["bWendu"] - df["yWendu"]

In [53]:

df.head()

Out[53]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

wencha

2018-01-01

-6

晴~多云

东北风

1-2级

良

2018-01-02

-5

阴~多云

东北风

1-2级

优

2018-01-03

-5

多云

北风

1-2级

优

2018-01-04

-8

阴

东北风

1-2级

优

2018-01-05

-6

多云~晴

西北风

1-2级

优

2. df.apply方法

Apply a function along an axis of the DataFrame. Objects passed to the function are Series objects whose index is either the DataFrame’s index (axis=0) or the DataFrame’s columns (axis=1) 实例：添加一列温度类型：

如果最高温度大于33度就是高温

低于-10度是低温

否则是常温

In [60]:

def get_wendu_type(x):

if x["bWendu"] > 33:

return "高温"

if x["yWendu"] < -10:

return "低温"

return "常温"

# 注意需要设置axis=1

df.loc[:,"wendu_type"] = df.apply(get_wendu_type, axis=1)

In [61]:

# 查看温度类型的计数

df["wendu_type"].value_counts()

Out[61]:

常温 328

高温 29

低温 8

Name: wendu_type, dtype: int64

3. df.assign方法

Assign new columns to a DataFrame.

Returns a new object with all original columns in addtion to new ones.

实例：将温度从摄氏度变成华氏度

In [63]:

# 可以同时添加多个新的列

df.assign(

yWendu_huashi = lambda x: x["yWendu"]*9/5 + 32,

bWendu_huashi = lambda x: x["bWendu"]*9/5 + 32

)

. . .

4. 按条件选择分组分别赋值

按条件选择数据，然后对整个数据赋值新列

实例：高低温差大于10度，则认为温差大

In [65]:

df.loc[:,"wencha_type"] = ""

df.loc[df["bWendu"]-df["yWendu"]>10, "wencha_type"] = "温差大"

df.loc[df["bWendu"]-df["yWendu"]<=10, "wencha_type"]= "温度正常"

In [67]:

df["wencha_type"].value_counts()

Out[67]:

温度正常 187

温差大 178

Name: wencha_type, dtype: int64

五、Pandas数据统计函数

汇总类统计

唯一去重和按值计数

tags

2585

Tensorflow怎样接收变长列表特征

python,tensorflow,特征工程

2583

Pandas实现数据的合并concat

pandas,python,数据分析

2574

Pandas的Index索引有什么用途？

pandas,python,数据分析

2564

机器学习常用数据集大全

python,机器学习

2561

一个数据科学家的修炼路径

数据分析

In [56]:

df_source.index

Out[56]:

RangeIndex(start=0, stop=258, step=1)

In [57]:

# 258行，3列

df_source.shape

Out[57]:

(258, 3)

In [58]:

# 通过df_source.shape得到元组(258, 3)

# 通过df_source.shape[0]得到行数

total_row_count = df_source.shape[0]

total_row_count

Out[58]:

258

1、将一个大excel等份拆成多个Excel

使用df.iloc方法，将一个大的DataFrame，拆分成多个小DataFrame

将使用DataFrame.to_excel保存每个小Excel

1. 1 计算拆分后的每个excel的行数

In [59]:

# 将一个大Excel，拆分给这几个人

user_names = ["A", "B", "C", "D", "E", "F"]

In [60]:

# 每个人的任务数目

split_size = total_row_count // len(user_names)

# 此处的作用在于如果有余数，可以将未分配的行数，分配给前面几人，保证所有的行都分配出去

if total_row_count % len(user_names) != 0:

split_size += 1

split_size

Out[60]:

1.2 拆分成多个DataFrame

In [64]:

df_subs = []

for idx, user_name in enumerate(user_names):

# iloc的开始索引

begin = idx*split_size

# iloc的结束索引

end = begin + split_size

# 实现df按照iloc拆分

df_sub = df_source.iloc[begin:end]

# 将每个子df存入列表

df_subs.append((idx, user_name, df_sub))

df_subs[0][2].head(5)

Out[64]:

title

tags

2585

Tensorflow怎样接收变长列表特征

python,tensorflow,特征工程

2583

Pandas实现数据的合并concat

pandas,python,数据分析

2574

Pandas的Index索引有什么用途？

pandas,python,数据分析

2564

机器学习常用数据集大全

python,机器学习

2561

一个数据科学家的修炼路径

数据分析

In [65]:

df_subs[1][2].head(5)

Out[65]:

title

tags

2120

Zookeeper并不保证读取的是最新数据

zookeeper

2089

Mybatis源码解读-初始化过程详解

mybatis

2076

怎样借助Python爬虫给宝宝起个好名字

python,爬虫

2022

Mybatis源码解读-设计模式总结

mybatis,设计模式

2012

打工者心态、主人公意识、个人公司品牌

程序人生

1.3 将每个DataFrame存入excel

In [63]:

for idx, user_name, df_sub in df_subs:

file_name = f"{splits_dir}/spike_pandas_{idx}_{user_name}.xlsx"

df_sub.to_excel(file_name, index=False)

2、合并多个小Excel到一个大Excel

遍历文件夹，得到要合并的Excel文件列表

分别读取到DataFrame，给每个df添加一列用于标记来源

使用pd.concat进行df批量合并

将合并后的DataFrame输出到excel

2.1 遍历文件夹，得到要合并的Excel名称列表

In [66]:

import os

excel_names = []

# listdir返回指定目录下的所有文件和文件夹名称

for excel_name in os.listdir(splits_dir):

excel_names.append(excel_name)

excel_names

Out[66]:

['spike_pandas_0_A.xlsx',

'spike_pandas_1_B.xlsx',

'spike_pandas_2_C.xlsx',

'spike_pandas_3_D.xlsx',

'spike_pandas_4_E.xlsx',

'spike_pandas_5_F.xlsx']

2.2 分别读取到DataFrame

In [70]:

df_list = []

for excel_name in excel_names:

# 读取每个excel到df

excel_path = f"{splits_dir}/{excel_name}"

df_split = pd.read_excel(excel_path)

# 得到username，通过字符串切片

username = excel_name.replace("spike_pandas_", "").replace(".xlsx", "")[2:]

# print(username)

# 给df_split添加一列username

df_split["username"] = username

df_list.append(df_split)

2.3 使用pd.concat进行合并

In [71]:

df_merged = pd.concat(df_list)

In [72]:

df_merged.shape

Out[72]:

(258, 4)

In [74]:

df_merged.head()

Out[74]:

title

tags

username

2585

Tensorflow怎样接收变长列表特征

python,tensorflow,特征工程

2583

Pandas实现数据的合并concat

pandas,python,数据分析

2574

Pandas的Index索引有什么用途？

pandas,python,数据分析

2564

机器学习常用数据集大全

python,机器学习

2561

一个数据科学家的修炼路径

数据分析

In [76]:

df_merged["username"].value_counts()

Out[76]:

B 43

F 43

D 43

E 43

A 43

C 43

Name: username, dtype: int64

2.4 将合并后的DataFrame输出到Excel

In [77]:

df_merged.to_excel(f"{work_dir}/spike_pandas_merged.xlsx", index=False)

十五、Pandas怎样实现groupby分组统计

类似SQL：

select city,max(temperature) from city_weather group by city;

groupby：先对数据分组，然后在每个分组上应用聚合函数、转换函数

本次演示：

一、分组使用聚合函数做数据统计

二、遍历groupby的结果理解执行流程

三、实例分组探索天气数据

In [1]:

import pandas as pd

import numpy as np

# 加上这一句，能在jupyter notebook展示matplo图表

%matplotlib inline

In [4]:

df = pd.DataFrame({'A': ['foo', 'bar', 'foo', 'bar', 'foo', 'bar', 'foo', 'foo'],

'B': ['one', 'one', 'two', 'three', 'two', 'two', 'one', 'three'],

'C': np.random.randn(8),

'D': np.random.randn(8)})

Out[4]:

foo

one

-0.102369

0.042233

bar

one

1.552845

-0.623522

foo

two

0.770077

0.205682

bar

three

-1.989910

-0.617111

foo

two

1.230455

-0.422428

bar

two

-0.697516

-0.964579

foo

one

-0.939646

-0.414017

foo

three

0.763570

0.451086

1、分组使用聚合函数做数据统计

1.1 单个列groupby，查询所有数据列的统计

In [5]:

df.groupby("A").sum()

Out[5]:

bar

-1.134582

-2.205211

foo

1.722086

-0.137444

我们看到：

groupby中的’A’变成了数据的索引列

因为要统计sum，但B列不是数字，所以被自动忽略掉

1.2 多个列groupby，查询所有数据列的统计

In [6]:

# 以A,B为索引，查询C,D的平均值

df.groupby(["A", "B"]).mean()

Out[6]:

bar

one

1.552845

-0.623522

three

-1.989910

-0.617111

two

-0.697516

-0.964579

foo

one

-0.521008

-0.185892

three

0.763570

0.451086

two

1.000266

-0.108373

In [7]:

# 取消A.B作为索引

df.groupby(["A", "B"], as_index=False).mean()

Out[7]:

bar

one

1.552845

-0.623522

bar

three

-1.989910

-0.617111

bar

two

-0.697516

-0.964579

foo

one

-0.521008

-0.185892

foo

three

0.763570

0.451086

foo

two

1.000266

-0.108373

1.3 同时查看多种数据统计

In [8]:

df.groupby("A").agg([np.sum, np.mean, np.std])

Out[8]:

sum

mean

std

sum

mean

std

bar

-1.134582

-0.378194

1.792834

-2.205211

-0.735070

0.198786

foo

1.722086

0.344417

0.864635

-0.137444

-0.027489

0.385242

我们看到：列变成了多级索引

1.4 查看单列的结果数据统计

In [10]:

# 预过滤，性能更好

df.groupby("A")["C"].agg([np.sum, np.mean, np.std])

Out[10]:

sum

mean

std

bar

-1.134582

-0.378194

1.792834

foo

1.722086

0.344417

0.864635

In [9]:

# 方法2

df.groupby("A").agg([np.sum, np.mean, np.std])["C"]

Out[9]:

sum

mean

std

bar

-1.134582

-0.378194

1.792834

foo

1.722086

0.344417

0.864635

1.5 不同列使用不同的聚合函数

In [12]:

# 以字典的形式对不同的列使用不同的聚合函数

df.groupby("A").agg({"C":np.sum, "D":np.mean})

Out[12]:

bar

-1.134582

-0.735070

foo

1.722086

-0.027489

2、遍历groupby的结果理解执行流程

for循环可以直接遍历每个group

2.1 遍历单个列聚合的分组

In [13]:

g = df.groupby("A")

Out[13]:

In [16]:

Out[16]:

foo

one

-0.102369

0.042233

bar

one

1.552845

-0.623522

foo

two

0.770077

0.205682

bar

three

-1.989910

-0.617111

foo

two

1.230455

-0.422428

bar

two

-0.697516

-0.964579

foo

one

-0.939646

-0.414017

foo

three

0.763570

0.451086

In [15]:

for name,group in g:

print(name)

print(group)

print()

# name:bar and foo

# group:是两个DataFrame

bar

A B C D

1 bar one 1.552845 -0.623522

3 bar three -1.989910 -0.617111

5 bar two -0.697516 -0.964579

foo

A B C D

0 foo one -0.102369 0.042233

2 foo two 0.770077 0.205682

4 foo two 1.230455 -0.422428

6 foo one -0.939646 -0.414017

7 foo three 0.763570 0.451086

*可以获取单个分组的数据*

In [17]:

g.get_group("bar")

Out[17]:

bar

one

1.552845

-0.623522

bar

three

-1.989910

-0.617111

bar

two

-0.697516

-0.964579

2.2 遍历多个列聚合的分组

In [20]:

g = df.groupby(["A", "B"])

Out[20]:

In [21]:

for name, group in g:

print(name)

print(group)

print()

# 分组的名称变成了元组

('bar', 'one')

A B C D

1 bar one 1.552845 -0.623522

('bar', 'three')

A B C D

3 bar three -1.98991 -0.617111

('bar', 'two')

A B C D

5 bar two -0.697516 -0.964579

('foo', 'one')

A B C D

0 foo one -0.102369 0.042233

6 foo one -0.939646 -0.414017

('foo', 'three')

A B C D

7 foo three 0.76357 0.451086

('foo', 'two')

A B C D

2 foo two 0.770077 0.205682

4 foo two 1.230455 -0.422428

可以看到，name是一个2个元素的tuple，代表不同的列

In [22]:

g.get_group(("foo", "one"))

Out[22]:

foo

one

-0.102369

0.042233

foo

one

-0.939646

-0.414017

*可以直接查询group后的某几列，生成Series或者子DataFrame*

In [24]:

# 获得一个SeriesGroupBy

g["C"]

Out[24]:

In [25]:

for name, group in g["C"]:

print(name)

print(group)

print(type(group))

print()

('bar', 'one')

1 1.552845

Name: C, dtype: float64

('bar', 'three')

3 -1.98991

Name: C, dtype: float64

('bar', 'two')

5 -0.697516

Name: C, dtype: float64

('foo', 'one')

0 -0.102369

6 -0.939646

Name: C, dtype: float64

('foo', 'three')

7 0.76357

Name: C, dtype: float64

('foo', 'two')

2 0.770077

4 1.230455

Name: C, dtype: float64

其实所有的聚合统计，都是在dataframe和series上进行的；

3、实例分组探索天气数据

In [27]:

fpath = "./pandas-learn-code/datas/beijing_tianqi/beijing_tianqi_2018.csv"

df = pd.read_csv(fpath)

df.head()

Out[27]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

2018-01-01

3℃

-6℃

晴~多云

东北风

1-2级

良

2018-01-02

2℃

-5℃

阴~多云

东北风

1-2级

优

2018-01-03

2℃

-5℃

多云

北风

1-2级

优

2018-01-04

0℃

-8℃

阴

东北风

1-2级

优

2018-01-05

3℃

-6℃

多云~晴

西北风

1-2级

优

In [28]:

# 替换掉温度的后缀℃

df.loc[:, "bWendu"] = df["bWendu"].str.replace("℃","").astype("int32")

df.loc[:, "yWendu"] = df["yWendu"].str.replace("℃","").astype("int32")

df.head()

Out[28]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

2018-01-01

-6

晴~多云

东北风

1-2级

良

2018-01-02

-5

阴~多云

东北风

1-2级

优

2018-01-03

-5

多云

北风

1-2级

优

2018-01-04

-8

阴

东北风

1-2级

优

2018-01-05

-6

多云~晴

西北风

1-2级

优

In [29]:

# 新增一列为月份

df["month"] = df["ymd"].str[:7]

df.head()

Out[29]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

month

2018-01-01

-6

晴~多云

东北风

1-2级

良

2018-01

2018-01-02

-5

阴~多云

东北风

1-2级

优

2018-01

2018-01-03

-5

多云

北风

1-2级

优

2018-01

2018-01-04

-8

阴

东北风

1-2级

优

2018-01

2018-01-05

-6

多云~晴

西北风

1-2级

优

2018-01

3.1 查看每个月的最高温度

In [31]:

data = df.groupby("month")["bWendu"].max()

data

Out[31]:

month

2018-01 7

2018-02 12

2018-03 27

2018-04 30

2018-05 35

2018-06 38

2018-07 37

2018-08 36

2018-09 31

2018-10 25

2018-11 18

2018-12 10

Name: bWendu, dtype: int32

In [32]:

type(data)

Out[32]:

pandas.core.series.Series

In [34]:

data.plot()

Out[34]:

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-Ssxk6ssS-1597761927710)(data:image/png;base64,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)]

3.2 查看每个月的最高温度、最低温度、平均空气质量指数

In [35]:

df.head()

Out[35]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

month

2018-01-01

-6

晴~多云

东北风

1-2级

良

2018-01

2018-01-02

-5

阴~多云

东北风

1-2级

优

2018-01

2018-01-03

-5

多云

北风

1-2级

优

2018-01

2018-01-04

-8

阴

东北风

1-2级

优

2018-01

2018-01-05

-6

多云~晴

西北风

1-2级

优

2018-01

In [38]:

group_data = df.groupby("month").agg({"bWendu":np.max, "yWendu":np.min, "aqi":np.mean})

group_data

Out[38]:

bWendu

yWendu

aqi

month

2018-01

-12

60.677419

2018-02

-10

78.857143

2018-03

-4

130.322581

2018-04

102.866667

2018-05

99.064516

2018-06

82.300000

2018-07

72.677419

2018-08

59.516129

2018-09

50.433333

2018-10

67.096774

2018-11

-4

105.100000

2018-12

-12

77.354839

In [39]:

group_data.plot()

Out[39]:

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-Lwmi8hwM-1597761927711)(data:image/png;base64,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)]

十六、Pandas的分层索引MultiIndex

为什么要学习分层索引MultiIndex？

分层索引：在一个轴向上拥有多个索引层级，可以表达更高维度数据的形式；

可以更方便的进行数据筛选，如果有序则性能更好；

groupby等操作的结果，如果是多KEY，结果是分层索引，需要会使用

一般不需要自己创建分层索引(MultiIndex有构造函数但一般不用)

演示数据：百度、阿里巴巴、爱奇艺、京东四家公司的10天股票数据

数据来自：英为财经

https://cn.investing.com/

本次演示提纲：

一、Series的分层索引MultiIndex

二、Series有多层索引怎样筛选数据？

三、DataFrame的多层索引MultiIndex

四、DataFrame有多层索引怎样筛选数据？

In [7]:

import pandas as pd

%matplotlib inline

In [8]:

fpath = "./pandas-learn-code/datas/stocks/互联网公司股票.xlsx"

stocks = pd.read_excel(fpath)

In [10]:

stocks.shape

Out[10]:

(12, 8)

In [5]:

stocks.head()

Out[5]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

In [12]:

stocks["公司"].unique()

Out[12]:

array(['BIDU', 'BABA', 'IQ', 'JD'], dtype=object)

In [14]:

# 按公司分组查询收盘价的平均值

stocks.groupby("公司")["收盘"].mean()

Out[14]:

公司

BABA 166.80

BIDU 102.98

IQ 15.90

JD 28.35

Name: 收盘, dtype: float64

1、Series的分层索引MultiIndex

In [16]:

# ser是Series,有两列索引

ser = stocks.groupby(["公司", "日期"])["收盘"].mean()

ser

Out[16]:

公司日期

BABA 2019-10-01 165.15

2019-10-02 165.77

2019-10-03 169.48

BIDU 2019-10-01 102.00

2019-10-02 102.62

2019-10-03 104.32

IQ 2019-10-01 15.92

2019-10-02 15.72

2019-10-03 16.06

JD 2019-10-01 28.19

2019-10-02 28.06

2019-10-03 28.80

Name: 收盘, dtype: float64

多维索引中，空白的意思是：使用上面的值

In [20]:

ser.index

Out[20]:

MultiIndex(levels=[['BABA', 'BIDU', 'IQ', 'JD'], ['2019-10-01', '2019-10-02', '2019-10-03']],

codes=[[0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3], [0, 1, 2, 0, 1, 2, 0, 1, 2, 0, 1, 2]],

names=['公司', '日期'])

In [21]:

# unstack把二级索引变成列

# 公司继续作为索引，但日期变为columns

ser.unstack()

Out[21]:

日期

2019-10-01

2019-10-02

2019-10-03

公司

BABA

165.15

165.77

169.48

BIDU

102.00

102.62

104.32

15.92

15.72

16.06

28.19

28.06

28.80

In [22]:

ser

Out[22]:

公司日期

BABA 2019-10-01 165.15

2019-10-02 165.77

2019-10-03 169.48

BIDU 2019-10-01 102.00

2019-10-02 102.62

2019-10-03 104.32

IQ 2019-10-01 15.92

2019-10-02 15.72

2019-10-03 16.06

JD 2019-10-01 28.19

2019-10-02 28.06

2019-10-03 28.80

Name: 收盘, dtype: float64

In [24]:

# 将两层索引(公司，日期)都变成了columns

ser.reset_index()

Out[24]:

公司

日期

收盘

BABA

2019-10-01

165.15

BABA

2019-10-02

165.77

BABA

2019-10-03

169.48

BIDU

2019-10-01

102.00

BIDU

2019-10-02

102.62

BIDU

2019-10-03

104.32

2019-10-01

15.92

2019-10-02

15.72

2019-10-03

16.06

2019-10-01

28.19

2019-10-02

28.06

2019-10-03

28.80

2、Series有多层索引MultiIndex怎么筛选数据？

In [25]:

ser

Out[25]:

公司日期

BABA 2019-10-01 165.15

2019-10-02 165.77

2019-10-03 169.48

BIDU 2019-10-01 102.00

2019-10-02 102.62

2019-10-03 104.32

IQ 2019-10-01 15.92

2019-10-02 15.72

2019-10-03 16.06

JD 2019-10-01 28.19

2019-10-02 28.06

2019-10-03 28.80

Name: 收盘, dtype: float64

In [27]:

ser.loc["BIDU"]

Out[27]:

日期

2019-10-01 102.00

2019-10-02 102.62

2019-10-03 104.32

Name: 收盘, dtype: float64

In [ ]:

# 多层索引，可以用元组的形式筛选

In [28]:

ser.loc[("BIDU","2019-10-02")]

Out[28]:

102.62

In [29]:

ser.loc[:, "2019-10-02"]

Out[29]:

公司

BABA 165.77

BIDU 102.62

IQ 15.72

JD 28.06

Name: 收盘, dtype: float64

3、DataFrame的多层索引MultiIndex

In [30]:

stocks.head()

Out[30]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

In [40]:

stocks.set_index(["公司", "日期"], inplace=True)

. . .

In [41]:

stocks.head()

Out[41]:

收盘

开盘

高

低

交易量

涨跌幅

公司

日期

BIDU

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-02

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-01

102.00

102.80

103.26

101.00

1.78

-0.01

BABA

2019-10-03

169.48

166.65

170.18

165.00

10.39

0.02

2019-10-02

165.77

162.82

166.88

161.90

11.60

0.00

In [42]:

stocks.index

Out[42]:

MultiIndex(levels=[['BABA', 'BIDU', 'IQ', 'JD'], ['2019-10-01', '2019-10-02', '2019-10-03']],

codes=[[1, 1, 1, 0, 0, 0, 2, 2, 2, 3, 3, 3], [2, 1, 0, 2, 1, 0, 2, 1, 0, 2, 1, 0]],

names=['公司', '日期'])

In [43]:

stocks.sort_index(inplace=True)

In [44]:

stocks

Out[44]:

收盘

开盘

高

低

交易量

涨跌幅

公司

日期

BABA

2019-10-01

165.15

168.01

168.23

163.64

14.19

-0.01

2019-10-02

165.77

162.82

166.88

161.90

11.60

0.00

2019-10-03

169.48

166.65

170.18

165.00

10.39

0.02

BIDU

2019-10-01

102.00

102.80

103.26

101.00

1.78

-0.01

2019-10-02

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-01

15.92

16.14

16.22

15.50

11.65

-0.01

2019-10-02

15.72

15.85

15.87

15.12

8.10

-0.01

2019-10-03

16.06

15.71

16.38

15.32

10.08

0.02

2019-10-01

28.19

28.22

28.57

27.97

10.64

0.00

2019-10-02

28.06

28.00

28.22

27.53

9.53

0.00

2019-10-03

28.80

28.11

28.97

27.82

8.77

0.03

4、DataFrame有多层索引MultiIndex怎样筛选？

【*重要知识*】在选择数据时：

元组(key1,key2)代表筛选多层索引，其中key1是索引第一级，key2是第二级，比如key1=JD, key2=2019-10-02

列表[key1,key2]代表同一层的多个KEY，其中key1和key2是并列的同级索引，比如key1=JD, key2=BIDU

In [45]:

stocks.loc["BIDU"]

Out[45]:

收盘

开盘

高

低

交易量

涨跌幅

日期

2019-10-01

102.00

102.80

103.26

101.00

1.78

-0.01

2019-10-02

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

In [46]:

# BIDU, 2019-10-02当天所有的相关数据

stocks.loc[("BIDU", "2019-10-02"), :]

Out[46]:

收盘 102.62

开盘 100.85

高 103.24

低 99.50

交易量 2.69

涨跌幅 0.01

Name: (BIDU, 2019-10-02), dtype: float64

In [48]:

# 逻辑关系为BIDU的2019-10-02的开盘数据

stocks.loc[("BIDU", "2019-10-02"), "开盘"]

Out[48]:

100.85

In [50]:

# 并列筛选，BIDU和JD为同级关系

stocks.loc[["BIDU", "JD"], :]

Out[50]:

收盘

开盘

高

低

交易量

涨跌幅

公司

日期

BIDU

2019-10-01

102.00

102.80

103.26

101.00

1.78

-0.01

2019-10-02

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-01

28.19

28.22

28.57

27.97

10.64

0.00

2019-10-02

28.06

28.00

28.22

27.53

9.53

0.00

2019-10-03

28.80

28.11

28.97

27.82

8.77

0.03

In [51]:

stocks.loc[(["BIDU", "JD"], "2019-10-03"), :]

Out[51]:

收盘

开盘

高

低

交易量

涨跌幅

公司

日期

BIDU

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-03

28.80

28.11

28.97

27.82

8.77

0.03

In [52]:

stocks.loc[(["BIDU", "JD"], "2019-10-03"), "收盘"]

Out[52]:

公司日期

BIDU 2019-10-03 104.32

JD 2019-10-03 28.80

Name: 收盘, dtype: float64

In [54]:

stocks.loc[("BIDU",["2019-10-02", "2019-10-03"]), "收盘"]

Out[54]:

公司日期

BIDU 2019-10-02 102.62

2019-10-03 104.32

Name: 收盘, dtype: float64

In [55]:

# slice(None)代表筛选这一索引的所有内容

stocks.loc[(slice(None), ["2019-10-02", "2019-10-03"]),:]

Out[55]:

收盘

开盘

高

低

交易量

涨跌幅

公司

日期

BABA

2019-10-02

165.77

162.82

166.88

161.90

11.60

0.00

2019-10-03

169.48

166.65

170.18

165.00

10.39

0.02

BIDU

2019-10-02

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-02

15.72

15.85

15.87

15.12

8.10

-0.01

2019-10-03

16.06

15.71

16.38

15.32

10.08

0.02

2019-10-02

28.06

28.00

28.22

27.53

9.53

0.00

2019-10-03

28.80

28.11

28.97

27.82

8.77

0.03

In [56]:

# 将多层索引恢复成列

stocks.reset_index()

Out[56]:

公司

日期

收盘

开盘

高

低

交易量

涨跌幅

BABA

2019-10-01

165.15

168.01

168.23

163.64

14.19

-0.01

BABA

2019-10-02

165.77

162.82

166.88

161.90

11.60

0.00

BABA

2019-10-03

169.48

166.65

170.18

165.00

10.39

0.02

BIDU

2019-10-01

102.00

102.80

103.26

101.00

1.78

-0.01

BIDU

2019-10-02

102.62

100.85

103.24

99.50

2.69

0.01

BIDU

2019-10-03

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-01

15.92

16.14

16.22

15.50

11.65

-0.01

2019-10-02

15.72

15.85

15.87

15.12

8.10

-0.01

2019-10-03

16.06

15.71

16.38

15.32

10.08

0.02

2019-10-01

28.19

28.22

28.57

27.97

10.64

0.00

2019-10-02

28.06

28.00

28.22

27.53

9.53

0.00

2019-10-03

28.80

28.11

28.97

27.82

8.77

0.03

十七、Pandas的数据转换函数map、apply、applymap

数据转换函数对比：map、apply、applymap：

map：只用于Series，实现每个值->值的映射；

apply：用于Series实现每个值的处理，用于Dataframe实现某个轴的Series的处理；

applymap：只能用于DataFrame，用于处理该DataFrame的每个元素；

1. map用于Series值的转换

实例：将股票代码英文转换成中文名字

Series.map(dict) or Series.map(function)均可

In [2]:

import pandas as pd

stocks = pd.read_excel(r"D:\WinterIsComing\python\New_Wave\pandas_basic\pandas-learn-code\datas\stocks\互联网公司股票.xlsx")

In [3]:

stocks.head()

Out[3]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

In [7]:

stocks["公司"].unique()

Out[7]:

array(['BIDU', 'BABA', 'IQ', 'JD'], dtype=object)

In [8]:

# 公司股票代码到中文的映射，注意这里是小写

dict_company_names={

"bidu":"百度",

"baba":"阿里巴巴",

"iq":"爱奇艺",

"jd":"京东"

}

方法1：Series.map(dict)

In [9]:

stocks["中文公司1"]=stocks["公司"].str.lower().map(dict_company_names)

In [10]:

stocks

Out[10]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

中文公司1

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

百度

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

百度

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

百度

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

阿里巴巴

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

阿里巴巴

2019-10-01

BABA

165.15

168.01

168.23

163.64

14.19

-0.01

阿里巴巴

2019-10-03

16.06

15.71

16.38

15.32

10.08

0.02

爱奇艺

2019-10-02

15.72

15.85

15.87

15.12

8.10

-0.01

爱奇艺

2019-10-01

15.92

16.14

16.22

15.50

11.65

-0.01

爱奇艺

2019-10-03

28.80

28.11

28.97

27.82

8.77

0.03

京东

2019-10-02

28.06

28.00

28.22

27.53

9.53

0.00

京东

2019-10-01

28.19

28.22

28.57

27.97

10.64

0.00

京东

方法2：Series.map(function)

function的参数是Series的每个元素的值

In [13]:

# lambda x中的x代表Series的每个值(即stocks["公司"]中的每个值)

stocks["公司中文2"]=stocks["公司"].map(lambda x : dict_company_names[x.lower()])

In [12]:

stocks.head()

Out[12]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

中文公司1

公司中文2

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

百度

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

百度

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

百度

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

阿里巴巴

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

阿里巴巴

2. apply用于Series和DataFrame的转换

Series.apply(function), 函数的参数是每个值

DataFrame.apply(function), 函数的参数是Series

Series.apply(function)

function的参数是Series的每个值

In [14]:

stocks["中文公司3"]=stocks["公司"].apply(lambda x : dict_company_names[x.lower()])

In [16]:

stocks.head()

Out[16]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

中文公司1

公司中文2

中文公司3

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

百度

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

百度

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

百度

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

阿里巴巴

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

阿里巴巴

DataFrame.apply(function)

function的参数是对应轴的Series

In [18]:

stocks["中文公司4"]=stocks.apply(lambda x: dict_company_names[x["公司"].lower()], axis=1)

In [19]:

stocks["公司"]

Out[19]:

0 BIDU

1 BIDU

2 BIDU

3 BABA

4 BABA

5 BABA

6 IQ

7 IQ

8 IQ

9 JD

10 JD

11 JD

Name: 公司, dtype: object

注意这个代码：

1、apply是在stocks这个DataFrame上调用；

2、lambda x的x是一个Series，因为指定了axis=1所以Seires的key是列名，可以用x[‘公司’]获取

In [20]:

stocks.head()

Out[20]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

中文公司1

公司中文2

中文公司3

中文公司4

2019-10-03

BIDU

104.32

102.35

104.73

101.15

2.24

0.02

百度

2019-10-02

BIDU

102.62

100.85

103.24

99.50

2.69

0.01

百度

2019-10-01

BIDU

102.00

102.80

103.26

101.00

1.78

-0.01

百度

2019-10-03

BABA

169.48

166.65

170.18

165.00

10.39

0.02

阿里巴巴

2019-10-02

BABA

165.77

162.82

166.88

161.90

11.60

0.00

阿里巴巴

3. applymap用于DataFrame所有值的转换

In [21]:

sub_df = stocks[["收盘","开盘","高","低","交易量"]]

In [22]:

sub_df

Out[22]:

收盘

开盘

高

低

交易量

104.32

102.35

104.73

101.15

2.24

102.62

100.85

103.24

99.50

2.69

102.00

102.80

103.26

101.00

1.78

169.48

166.65

170.18

165.00

10.39

165.77

162.82

166.88

161.90

11.60

165.15

168.01

168.23

163.64

14.19

16.06

15.71

16.38

15.32

10.08

15.72

15.85

15.87

15.12

8.10

15.92

16.14

16.22

15.50

11.65

28.80

28.11

28.97

27.82

8.77

28.06

28.00

28.22

27.53

9.53

28.19

28.22

28.57

27.97

10.64

In [23]:

# 将这些数字取整数，应用于所有元素(即表格中所有的值)

sub_df.applymap(lambda x: int(x))

Out[23]:

收盘

开盘

高

低

交易量

104

102

104

101

102

100

103

102

103

101

169

166

170

165

162

166

161

165

168

163

In [25]:

# 直接修改原df的这几列

stocks.loc[:, ["收盘","开盘","高","低","交易量"]] = sub_df.applymap(lambda x: int(x))

In [26]:

stocks.head()

Out[26]:

日期

公司

收盘

开盘

高

低

交易量

涨跌幅

中文公司1

公司中文2

中文公司3

中文公司4

2019-10-03

BIDU

104

102

104

101

0.02

百度

2019-10-02

BIDU

102

100

103

0.01

百度

2019-10-01

BIDU

102

103

101

-0.01

百度

2019-10-03

BABA

169

166

170

165

0.02

阿里巴巴

2019-10-02

BABA

165

162

166

161

0.00

阿里巴巴

十八、Pandas怎样对每个分组应用apply函数?

知识：Pandas的GroupBy遵从split、apply、combine模式

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-lUn7YSnp-1597761927713)(http://localhost:8891/notebooks/pandas-learn-code/other_files/pandas-split-apply-combine.png)]

这里的split指的是pandas的groupby，我们自己实现apply函数，apply返回的结果由pandas进行combine得到结果

GroupBy.apply(function)

function的第一个参数是dataframe

function的返回结果，可是dataframe、series、单个值，甚至和输入dataframe完全没关系

本次实例演示：

怎样对数值列按分组的归一化？

怎样取每个分组的TOPN数据？

实例1：怎样对数值列按分组的归一化？

将不同范围的数值列进行归一化，映射到[0,1]区间：

更容易做数据横向对比，比如价格字段是几百到几千，增幅字段是0到100

机器学习模型学的更快性能更好

归一化的公式：

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-CuwTEdIl-1597761927714)(http://localhost:8891/notebooks/pandas-learn-code/other_files/Normalization-Formula.jpg)]

演示：用户对电影评分的归一化

每个用户的评分不同，有的乐观派评分高，有的悲观派评分低，按用户做归一化

In [1]:

import pandas as pd

In [7]:

ratings = pd.read_csv(

"./pandas-learn-code/datas/movielens-1m/ratings.dat",

sep="::",

engine="python",

names="UserID::MovieID::Rating::Timestamp".split("::")

)

In [8]:

ratings.head()

Out[8]:

UserID

MovieID

Rating

Timestamp

1193

978300760

661

978302109

914

978301968

3408

978300275

2355

978824291

In [10]:

# 实现按照用户ID分组，然后对ratings进行归一化

def ratings_norm(df):

# 实际参数是每个用户分组的df(按照UserID分组的DataFrame)

max_value = df["Rating"].max()

min_value = df["Rating"].min()

df["Rating_norm"] = df["Rating"].apply(lambda x:(x - min_value)/(max_value - min_value))

return df

# 按照用户分组，apply一个函数，给该DataFrame新增了一列，实现了Rating列的归一化

ratings = ratings.groupby("UserID").apply(ratings_norm)

In [12]:

ratings["Rating"]

. . .

In [16]:

type(ratings)

Out[16]:

pandas.core.frame.DataFrame

In [17]:

ratings[ratings["UserID"]==1].head()

Out[17]:

UserID

MovieID

Rating

Timestamp

Rating_norm

1193

978300760

1.0

661

978302109

0.0

914

978301968

0.0

3408

978300275

0.5

2355

978824291

1.0

实例2：怎么取每个分组的TOP N数据

获取2018年每个月温度最高的2天数据

In [18]:

fpath = "./pandas-learn-code/datas/beijing_tianqi/beijing_tianqi_2018.csv"

df = pd.read_csv(fpath)

In [19]:

df.head()

Out[19]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

2018-01-01

3℃

-6℃

晴~多云

东北风

1-2级

良

2018-01-02

2℃

-5℃

阴~多云

东北风

1-2级

优

2018-01-03

2℃

-5℃

多云

北风

1-2级

优

2018-01-04

0℃

-8℃

阴

东北风

1-2级

优

2018-01-05

3℃

-6℃

多云~晴

西北风

1-2级

优

In [21]:

# 替换掉温度后的℃

df.loc[:, "bWendu"]=df["bWendu"].str.replace("℃","").astype("int32")

df.loc[:, "yWendu"]=df["yWendu"].str.replace("℃","").astype("int32")

In [22]:

# 新增一列为月份

df["month"] = df["ymd"].str[0:7]

df.head()

Out[22]:

ymd

bWendu

yWendu

tianqi

fengxiang

fengli

aqi

aqiInfo

aqiLevel

month

2018-01-01

-6

晴~多云

东北风

1-2级

良

2018-01

2018-01-02

-5

阴~多云

东北风

1-2级

优

2018-01

2018-01-03

-5

多云

北风

1-2级

优

2018-01

2018-01-04

-8

阴

东北风

1-2级

优

2018-01

2018-01-05

-6

多云~晴

西北风

1-2级

优

2018-01

In [24]:

def getWenduTopN(df, topn):

# 这里的df，是每个月份分组group的df

return df.sort_values(by="bWendu")[["ymd", "bWendu"]][-topn:]

df.groupby("month").apply(getWenduTopN, topn=2).head()

Out[24]:

ymd

bWendu

month

2018-01

2018-01-14

2018-01-19

2018-02

2018-02-23

2018-02-26

2018-03

2018-03-28

In [25]:

df[["ymd","bWendu"]]

. . .

我们看到，groupby的apply函数返回的DataFrame，其实和原来的DataFrame其实可以完全不一样

php正则匹配sg-nc-wap_Pandas基础操作（上）

你可能感兴趣的:(php正则匹配sg-nc-wap_Pandas基础操作（上）)