import pandas as pd
import numpy as np
from pandas import Series,DataFrame
data = {
'k1':['a','a','b','b','b'],
'k2':['one','two','one','two','one'],
'data1':range(5),
'data2':np.random.randn(5)
}
df = DataFrame(data) #data1 data2 k1 k2
#0 0 -1.751940 a one
#1 1 0.355081 a two
#2 2 -1.105436 b one
#3 3 0.839877 b two
#4 4 -0.761272 b one
#一、分组键为Series
#根据k1键进行分组
grouped = df['data1'].groupby(df['k1']) #<pandas.core.groupby.SeriesGroupBy object at 0x000001FA54483F98>
#求每组的平均值
grouped.mean() #k1
#a 0.5
#b 3.0
#根据k1,k2键进行分组
#得到一个层次化的索引
df['data1'].groupby([df['k1'],df['k2']]).mean() #k1 k2
#a one 0
# two 1
#b one 3
# two 3
df['data1'].groupby([df['k1'],df['k2']]).mean().unstack() #k2 one two
#k1
#a 0 1
#b 3 3
#二、分组键为长度适当的任意数组
#根据一个自定义数组进行分组（分组键可以是长度适当的任意数组）
years=np.array([2005,2006,2006,2006,2005])
df['data1'].groupby(years).mean() #2005 2
#2006 2
#三、分组键为列名
df.groupby(['k1']).mean() #data1 data2
#k1
#a 0.5 0.364062
#b 3.0 0.124075
#四、groupby的size方法，返回一个有分组大小的Series
df.groupby(['k1']).size() #k1
#a 2
#b 3
#五、对分组进行迭代(分组产生一个二元元组，由分组名和数据块组成)
#一个分组键的情况
#for name,group in df.groupby('k1'):
#print(name)
#print(group) #a
# data1 data2 k1 k2
#0 0 0.104821 a one
#1 1 0.177355 a two
#b
# data1 data2 k1 k2
#2 2 -0.676046 b one
#3 3 0.806155 b two
#4 4 -0.506097 b one
#多个分组键的情况,元组的第一个元素是由分组键组成的元组
#for (k1,k2),group in df.groupby(['k1','k2']):
#print(k1,k2)
#print(group) #a one
# data1 data2 k1 k2
#0 0 -0.879509 a one
#a two
# data1 data2 k1 k2
#1 1 -0.309208 a two
#b one
# data1 data2 k1 k2
#2 2 0.006868 b one
#4 4 -0.736024 b one
#b two
# data1 data2 k1 k2
#3 3 -0.612376 b two
#可以看出分组对象是一个二元元组，由分组键和数据块组成
list(df.groupby('k1')) #[('a', data1 data2 k1 k2
# 0 0 0.296102 a one
# 1 1 0.761831 a two), ('b', data1 data2 k1 k2
# 2 2 0.763685 b one
# 3 3 -0.105020 b two
# 4 4 -0.483341 b one)]
#转换成字典
pieces = dict(list(df.groupby('k1')))
pieces['a'] # data1 data2 k1 k2
# 0 0 -0.142410 a one
# 1 1 1.059362 a two
pieces['b'] # data1 data2 k1 k2
# 2 2 0.865926 b one
# 3 3 1.479096 b two
# 4 4 0.668402 b one
#六、通过对groupby对象进行索引,能达到选取指定列 df.groupby('k1')['data1']等同 df['data1'].groupby('k1')
df.groupby('k1')['data1']
#七、通过字典进行分组
data = DataFrame(np.arange(20).reshape(4,5),columns=['a','b','c','d','e'])
data #a b c d e
#0 0 1 2 3 4
#1 5 6 7 8 9
#2 10 11 12 13 14
#3 15 16 17 18 19
mapping={'a':'一组','b':'一组','c':'一组','d':'二组','e':'二组'}
data.groupby(mapping,axis=1).sum() # 一组 二组
#0 3 7
#1 18 17
#2 33 27
#3 48 37
#八、通过函数进行分组
data = DataFrame(np.arange(20).reshape(4,5),columns=['a','b','c','d','e'],index=['bob','tom','lily','joee'])
data # a b c d e
#bob 0 1 2 3 4
#tom 5 6 7 8 9
#lily 10 11 12 13 14
#joee 15 16 17 18 19
data.groupby(len).sum() #a b c d e
#3 5 7 9 11 13
#4 25 27 29 31 33