透视聚合
import pandas as pd
import numpy as np
df = pd.DataFrame({'Name':['Mary', 'Jon','Lucy', 'Jane', 'Sue', 'Mary', 'Lucy'],
'Age':[35, 37, 40, 29, 31, 26, 28],
'City':['Boston', 'Chicago', 'Los Angeles', 'Chicago', 'Boston', 'Boston', 'Chicago'],
'Position':['Manager','Manager','Manager','Programmer', 'Programmer','Manager','Manager'],
'Sex':['Female','Male','Female','Female', 'Female','Female','Female']},
columns=['Name','Position','City','Age','Sex'])
print (df)
Name Position City Age Sex
0 Mary Manager Boston 35 Female
1 Jon Manager Chicago 37 Male
2 Lucy Manager Los Angeles 40 Female
3 Jane Programmer Chicago 29 Female
4 Sue Programmer Boston 31 Female
5 Mary Manager Boston 26 Female
6 Lucy Manager Chicago 28 Female
如果使用 pivot ，得到错误：
print (df.pivot(index='Position', columns='City', values='Age'))
ValueError：索引包含重复的条目，无法重新整形
使用 pivot_table 和聚合功能：
#default aggfunc is np.mean
print (df.pivot_table(index='Position', columns='City', values='Age'))
City Boston Chicago Los Angeles
Position
Manager 30.5 32.5 40.0
Programmer 31.0 29.0 NaN
print (df.pivot_table(index='Position', columns='City', values='Age', aggfunc=np.mean))
City Boston Chicago Los Angeles
Position
Manager 30.5 32.5 40.0
Programmer 31.0 29.0 NaN
另一个 agg 功能：
print (df.pivot_table(index='Position', columns='City', values='Age', aggfunc=sum))
City Boston Chicago Los Angeles
Position
Manager 61.0 65.0 40.0
Programmer 31.0 29.0 NaN
#lost data !!!
print (df.pivot_table(index='Position', columns='City', values='Age', aggfunc='first'))
City Boston Chicago Los Angeles
Position
Manager 35.0 37.0 40.0
Programmer 31.0 29.0 NaN
如果需要按 string 值的列聚合：
print (df.pivot_table(index='Position', columns='City', values='Name'))
DataError：没有要聚合的数字类型
你可以使用这些 aggragating 功能：
print (df.pivot_table(index='Position', columns='City', values='Name', aggfunc='first'))
City Boston Chicago Los Angeles
Position
Manager Mary Jon Lucy
Programmer Sue Jane None
print (df.pivot_table(index='Position', columns='City', values='Name', aggfunc='last'))
City Boston Chicago Los Angeles
Position
Manager Mary Lucy Lucy
Programmer Sue Jane None
print (df.pivot_table(index='Position', columns='City', values='Name', aggfunc='sum'))
City Boston Chicago Los Angeles
Position
Manager MaryMary JonLucy Lucy
Programmer Sue Jane None
print (df.pivot_table(index='Position', columns='City', values='Name', aggfunc=', '.join))
City Boston Chicago Los Angeles
Position
Manager Mary, Mary Jon, Lucy Lucy
Programmer Sue Jane None
print (df.pivot_table(index='Position', columns='City', values='Name', aggfunc=', '.join, fill_value='-')
.reset_index()
.rename_axis(None, axis=1))
Position Boston Chicago Los Angeles
0 Manager Mary, Mary Jon, Lucy Lucy
1 Programmer Sue Jane -
有关性别的信息尚未使用。它可以由其中一个列切换，也可以作为另一个级别添加：
print (df.pivot_table(index='Position', columns=['City','Sex'], values='Age', aggfunc='first'))
City Boston Chicago Los Angeles
Sex Female Female Male Female
Position
Manager 35.0 28.0 37.0 40.0
Programmer 31.0 29.0 NaN NaN
可以在索引，列和值的任何属性中指定多个列。
print (df.pivot_table(index=['Position','Sex'], columns='City', values='Age', aggfunc='first'))
City Boston Chicago Los Angeles
Position Sex
Manager Female 35.0 28.0 40.0
Male NaN 37.0 NaN
Programmer Female 31.0 29.0 NaN
应用几个聚合函数
你可以在单个数据透视中轻松应用多个功能：
In [23]: import numpy as np
In [24]: df.pivot_table(index='Position', values='Age', aggfunc=[np.mean, np.std])
Out[24]:
mean std
Position
Manager 34.333333 5.507571
Programmer 32.333333 4.163332
有时，你可能希望将特定函数应用于特定列：
In [35]: df['Random'] = np.random.random(6)
In [36]: df
Out[36]:
Name Position City Age Random
0 Mary Manager Boston 34 0.678577
1 Josh Programmer New York 37 0.973168
2 Jon Manager Chicago 29 0.146668
3 Lucy Manager Los Angeles 40 0.150120
4 Jane Programmer Chicago 29 0.112769
5 Sue Programmer Boston 31 0.185198
For example, find the mean age, and standard deviation of random by Position:
In [37]: df.pivot_table(index='Position', aggfunc={'Age': np.mean, 'Random': np.std})
Out[37]:
Age Random
Position
Manager 34.333333 0.306106
Programmer 32.333333 0.477219
可以传递一系列函数以应用于各个列：
In [38]: df.pivot_table(index='Position', aggfunc={'Age': np.mean, 'Random': [np.mean, np.std]})]
Out[38]:
Age Random
mean mean std
Position
Manager 34.333333 0.325122 0.306106
Programmer 32.333333 0.423712 0.477219