這篇groupby寫的不好。太復(fù)雜了。其實(shí)實(shí)際上經(jīng)常用的就那么幾個(gè)。舉個(gè)例子，把常用的往那一放就很容易理解和拿來用了。日后再寫一篇。

groupby默認(rèn)縱方向上分組，axis=0

DataFrame

import pandas as pd
import numpy as np

 df = pd.DataFrame({'key1':['a', 'a', 'b', 'b', 'a'],
                    'key2':['one', 'two', 'one', 'two', 'one'],
                    'data1':np.random.randn(5),
                    'data2':np.random.randn(5)})
print(df)

      data1     data2 key1 key2
0 -0.410122  0.247895    a  one
1 -0.627470 -0.989268    a  two
2  0.179488 -0.054570    b  one
3 -0.299878 -1.640494    b  two
4 -0.297191  0.954447    a  one

分組，并對(duì)分組進(jìn)行迭代

list(df.groupby(['key1']))#list后得到：[(group1),(group2),......]

[('a',       data1     data2 key1 key2
  0 -0.410122  0.247895    a  one
  1 -0.627470 -0.989268    a  two
  4 -0.297191  0.954447    a  one), ('b',       data1     data2 key1 key2
  2  0.179488 -0.054570    b  one
  3 -0.299878 -1.640494    b  two)]

list后得到：[(group1),(group2),…]
每個(gè)數(shù)據(jù)片(group)格式: (name,group)元組

1. 按key1（一個(gè)列）分組，其實(shí)是按key1的值

groupby對(duì)象支持迭代，產(chǎn)生一組二元元組：（分組名，數(shù)據(jù)塊）,（分組名，數(shù)據(jù)塊）…

for name,group in df.groupby(['key1']):
    print(name)
    print(group)

a
      data1     data2 key1 key2
0 -0.410122  0.247895    a  one
1 -0.627470 -0.989268    a  two
4 -0.297191  0.954447    a  one
b
      data1     data2 key1 key2
2  0.179488 -0.054570    b  one
3 -0.299878 -1.640494    b  two

2. 按[key1, key2]（多個(gè)列）分組

對(duì)于多重鍵，產(chǎn)生的一組二元元組：（（k1,k2），數(shù)據(jù)塊）,（（k1,k2），數(shù)據(jù)塊）…
第一個(gè)元素是由鍵值組成的元組

for name,group in df.groupby(['key1','key2']):
    print(name)  #name=(k1,k2)
    print(group)

('a', 'one')
      data1     data2 key1 key2
0 -0.410122  0.247895    a  one
4 -0.297191  0.954447    a  one
('a', 'two')
     data1     data2 key1 key2
1 -0.62747 -0.989268    a  two
('b', 'one')
      data1    data2 key1 key2
2  0.179488 -0.05457    b  one
('b', 'two')
      data1     data2 key1 key2
3 -0.299878 -1.640494    b  two

3. 按函數(shù)分組

4. 按字典分組

5. 按索引級(jí)別分組

6.將函數(shù)跟數(shù)組、列表、字典、Series混合使用也不是問題，因?yàn)槿魏螙|西最終都會(huì)被轉(zhuǎn)換為數(shù)組

將這些數(shù)據(jù)片段做成字典

dict(list(df.groupby(['key1'])))#dict(list())

{'a':       data1     data2 key1 key2
 0 -0.410122  0.247895    a  one
 1 -0.627470 -0.989268    a  two
 4 -0.297191  0.954447    a  one, 'b':       data1     data2 key1 key2
 2  0.179488 -0.054570    b  one
 3 -0.299878 -1.640494    b  two}

分組后進(jìn)行一些統(tǒng)計(jì)、計(jì)算等

1. 分組后，返回一個(gè)含有分組大小的Series

按key1分組

df.groupby(['key1']).size()

key1
a    3
b    2
dtype: int64

dict(['a1','x2','e3'])

{'a': '1', 'e': '3', 'x': '2'}

按[key1,key2]分組

df.groupby(['key1','key2']).size()

key1  key2
a     one     2
      two     1
b     one     1
      two     1
dtype: int64

2. 對(duì)data1按key1進(jìn)行分組，并計(jì)算data1列的平均值

df['data1'].groupby(df['key1']).mean()
#groupby沒有進(jìn)行任何的計(jì)算。它只是進(jìn)行了一個(gè)分組

key1
a   -0.444928
b   -0.060195
Name: data1, dtype: float64

df.groupby(['key1'])['data1'].mean()#理解：對(duì)df按key1分組，并計(jì)算分組后df['data1']的均值
#等價(jià)于：df.groupby(['key1']).data1.mean()

key1
a   -0.444928
b   -0.060195
Name: data1, dtype: float64

說明：
groupby沒有進(jìn)行任何的計(jì)算。它只是進(jìn)行了一個(gè)分組。
數(shù)據(jù)（Series）根據(jù)分組鍵進(jìn)行了聚合，產(chǎn)生了一個(gè)新的Series，其索引為key1列中的唯一值。

這種索引操作所返回的對(duì)象是一個(gè)已分組的DataFrame（如果傳入的是列表或數(shù)組）或已分組的Series

df.groupby(['key1'])['data1'].size()

key1
a    3
b    2
Name: data1, dtype: int64

3.對(duì)data1按[key1,key2]進(jìn)行分組，并計(jì)算data1的平均值

df['data1'].groupby([df['key1'],df['key2']]).mean()

key1  key2
a     one    -0.353657
      two    -0.627470
b     one     0.179488
      two    -0.299878
Name: data1, dtype: float64

df.groupby(['key1','key2'])['data1'].mean()
#等價(jià)于：df.groupby(['key1','key2']).data1'.mean()

key1  key2
a     one    -0.353657
      two    -0.627470
b     one     0.179488
      two    -0.299878
Name: data1, dtype: float64

通過兩個(gè)鍵對(duì)數(shù)據(jù)進(jìn)行了分組，得到的Series具有一個(gè)層次化索引（由唯一的鍵對(duì)組成）：

df.groupby(['key1','key2'])['data1'].mean().unstack()

在上面這些示例中，分組鍵均為Series。實(shí)際上，分組鍵可以是任何長(zhǎng)度適當(dāng)?shù)臄?shù)組。非常靈活。

橫方向上

按列的數(shù)據(jù)類型（df.dtypes）來分

df共兩種數(shù)據(jù)類型：float64和object，所以會(huì)分為兩組（dtype(‘float64’）,數(shù)據(jù)片),（dtype(‘O’), 數(shù)據(jù)片）

list(df.groupby(df.dtypes, axis=1))

[(dtype('float64'),       data1     data2
  0 -0.410122  0.247895
  1 -0.627470 -0.989268
  2  0.179488 -0.054570
  3 -0.299878 -1.640494
  4 -0.297191  0.954447), (dtype('O'),   key1 key2
  0    a  one
  1    a  two
  2    b  one
  3    b  two
  4    a  one)]

agg的應(yīng)用

groupby+agg 可以對(duì)groupby的結(jié)果同時(shí)應(yīng)用多個(gè)函數(shù)

SeriesGroupBy的方法agg()參數(shù)：
aggregate(self, func_or_funcs, * args, ** kwargs)
func: function, string, dictionary, or list of string/functions
返回：aggregated的Series

s= pd.Series([10,20,30,40])
s

0    10
1    20
2    30
3    40
dtype: int64

for n,g in s.groupby([1,1,2,2]):
    print(n)
    print(g)

1
0    10
1    20
dtype: int64
2
2    30
3    40
dtype: int64

s.groupby([1,1,2,2]).min()

1    10
2    30
dtype: int64

#等價(jià)于這個(gè)：
s.groupby([1,1,2,2]).agg('min')

1    10
2    30
dtype: int64

s.groupby([1,1,2,2]).agg(['min','max'])#加[]，func僅接受一個(gè)參數(shù)

常常這樣用：

df

比較下面，可以看出agg的用處：

df.groupby(['key1'])['data1'].min()

key1
a   -0.627470
b   -0.299878
Name: data1, dtype: float64

df.groupby(['key1'])['data1'].agg({'min'})

#推薦用這個(gè)√
df.groupby(['key1']).agg({'data1':'min'})#對(duì)data1列，取各組的最小值，名字還是data1

#按key1分組后，aggregate各組data1的最小值和最大值：
df.groupby(['key1'])['data1'].agg({'min','max'})

#推薦用這個(gè)√
df.groupby(['key1']).agg({'data1':['min','max']})

可以對(duì)groupby的結(jié)果更正列名（不推薦用這個(gè)，哪怕在后面單獨(dú)更改列名）

# 對(duì)data1，把min更名為a，max更名為b
df.groupby(['key1'])['data1'].agg({'a':'min','b':'max'})#這里的'min' 'max'為兩個(gè)函數(shù)名

d:\python27\lib\site-packages\ipykernel_launcher.py:2: FutureWarning: using a dict on a Series for aggregation
is deprecated and will be removed in a future version