2 重塑和轴向旋转
2.1 重塑层次化索引
针对DataFrame，有stack和unstack。stack将数据的列“旋转”为行，unstack反之。
In [6]: data = DataFrame(np.arange(6).reshape(2,3),index=pd.Index(['Ohio','Color
...: ado'], name='state'),columns=pd.Index(['one','two','three'],name='number
...: '))
In [7]: data
Out[7]:
number one two three
state
Ohio 0 1 2
Colorado 3 4 5
In [8]: data.stack() # 把列转化为行，即形成一个层次化索引的Series
Out[8]:
state number
Ohio one 0
two 1
three 2
Colorado one 3
two 4
three 5
dtype: int64
In [9]: data.stack().unstack() # 把层次化索引的Series转化为一个普通的DataFrame
Out[9]:
number one two three
state
Ohio 0 1 2
Colorado 3 4 5
In [10]: data.stack().unstack('number')
Out[10]:
number one two three
state
Ohio 0 1 2
Colorado 3 4 5
In [11]: data.stack().unstack(1)
Out[11]:
number one two three
state
Ohio 0 1 2
Colorado 3 4 5
In [12]: data.stack().unstack(0)
Out[12]:
state Ohio Colorado
number
one 0 3
two 1 4
three 2 5
In [13]: data.stack().unstack('state') # 选择指定的列，进行旋转
Out[13]:
state Ohio Colorado
number
one 0 3
two 1 4
three 2 5
对于Series也是一样，
In [14]: s1 = Series([0,1,2,3],index=list('abcd'))
In [15]: s2 = Series([4,5,6],index=list('cde'))
In [16]: data2 = pd.concat([s1,s2],keys=['one','two'])
In [17]: data2
Out[17]:
one a 0
b 1
c 2
d 3
two c 4
d 5
e 6
dtype: int64
In [18]: data2.unstack()
Out[18]:
a b c d e
one 0.0 1.0 2.0 3.0 NaN
two NaN NaN 4.0 5.0 6.0
In [19]: data2.unstack().stack() # 默认丢弃nan值
Out[19]:
one a 0.0
b 1.0
c 2.0
d 3.0
two c 4.0
d 5.0
e 6.0
dtype: float64
In [20]: data2.unstack().stack(dropna=False) # 不丢弃nan值
Out[20]:
one a 0.0
b 1.0
c 2.0
d 3.0
e NaN
two a NaN
b NaN
c 4.0
d 5.0
e 6.0
dtype: float64
2.2 将“长格式”转化为“宽格式”
时间序列的通常以“长格式”或“堆栈”格式存放在数据库或者CSV中。
In [28]: df = pd.DataFrame({'A' : ['one', 'one', 'two', 'three'] * 3,
...: 'B' : ['A', 'B', 'C'] * 4,'C' : ['foo', 'foo', 'foo', 'bar', 'bar', 'ba
...: r'] * 2,
...: 'D' : np.random.randn(12),'E' : np.random.randn(12)})
In [29]: df
Out[29]:
A B C D E
0 one A foo -1.130142 -0.800142
1 one B foo 1.453914 -1.344615
2 two C foo 0.452666 0.103939
3 three A bar 0.053393 0.284834
4 one B bar -0.435699 0.885828
5 one C bar 0.492411 -1.280743
6 two A foo -1.819792 -1.654636
7 three B foo 0.364982 1.654728
8 one C foo -1.205436 0.712150
9 one A bar 2.696329 1.357930
10 two B bar -0.561867 1.326672
11 three C bar 0.782410 -0.600689
# 这样就是df表中以D作为数值域，A,B为行，C为列的数据视图。
In [30]: pd.pivot_table(df, values='D', index=['A', 'B'], columns=['C'])
Out[30]:
C bar foo
A B
one A 2.696329 -1.130142
B -0.435699 1.453914
C 0.492411 -1.205436
three A 0.053393 NaN
B NaN 0.364982
C 0.782410 NaN
two A NaN -1.819792
B -0.561867 NaN
C NaN 0.452666
3 数据转换
前面两部分说的都是数据的重排，还有一类操作就是过滤、清理和转换。
3.1 移除重复数据
duplicated方法返回一个bool型Series，判断行是否为重复行。
drop_duplicates方法：返回一个删除了重复行的DataFrame
In [31]: data = DataFrame({'k1':['one']*3+['two']*4,'k2':[1,1,2,3,3,4,4]})
In [32]: data
Out[32]:
k1 k2
0 one 1
1 one 1
2 one 2
3 two 3
4 two 3
5 two 4
6 two 4
In [33]: data.duplicated()
Out[33]:
0 False
1 True
2 False
3 False
4 True
5 False
6 True
dtype: bool
In [34]: data.duplicated('k2') # 按照列进行判断
Out[34]:
0 False
1 True
2 False
3 False
4 True
5 False
6 True
dtype: bool
In [35]: data.drop_duplicates
Out[35]:
<bound method DataFrame.drop_duplicates of k1 k2
0 one 1
1 one 1
2 one 2
3 two 3
4 two 3
5 two 4
6 two 4>
In [36]: data.drop_duplicates() # 返回一个新的去重后的对象
Out[36]:
k1 k2
0 one 1
2 one 2
3 two 3
5 two 4
In [37]: data # 原对象不变
Out[37]:
k1 k2
0 one 1
1 one 1
2 one 2
3 two 3
4 two 3
5 two 4
6 two 4
# 默认保留的是第一个重复值
In [41]: data.drop_duplicates(['k1','k2']) #
Out[41]:
k1 k2 v1
0 one 1 0
2 one 2 2
3 two 3 3
5 two 4 5
In [42]: data.drop_duplicates(['k1','k2'],take_last=True)
/Users/yangfeilong/anaconda/bin/ipython:1: FutureWarning: the take_last=True keyword is deprecated, use keep='last' instead
#!/bin/bash /Users/yangfeilong/anaconda/bin/python.app
Out[42]:
k1 k2 v1
1 one 1 1
2 one 2 2
4 two 3 4
6 two 4 6
# 也可以指定保留最后一个值
In [43]: data.drop_duplicates(['k1','k2'],keep='last')
Out[43]:
k1 k2 v1
1 one 1 1
2 one 2 2
4 two 3 4
6 two 4 6
In [44]: data.drop_duplicates(['k1','k2'],keep='first')
Out[44]:
k1 k2 v1
0 one 1 0
2 one 2 2
3 two 3 3
5 two 4 5
3.2 利用函数或映射进行数据转换
映射可以是一个字典，key为源，value为目标。
如何转换呢？DataFrame通过apply或者applymap，Series通过map函数进行。
In [45]: data = DataFrame({'food':['bacon','pulled pork','bacon','Pastrami','cor
...: ned beef','Bacon','pastrami','honey ham','nova lox'],'ounces':[4,3,12,6
...: ,7.5,8,3,5,6]})
In [46]: data
Out[46]:
food ounces
0 bacon 4.0
1 pulled pork 3.0
2 bacon 12.0
3 Pastrami 6.0
4 corned beef 7.5
5 Bacon 8.0
6 pastrami 3.0
7 honey ham 5.0
8 nova lox 6.0
In [47]: meat_to_animal = {'bacon':'pig','pulled pork':'pig','pastrami':'cow','c
...: orned beef':'cow','honey ham':'pig','nova lox':'salmon'}
In [49]: data['food']
Out[49]:
0 bacon
1 pulled pork
2 bacon
3 Pastrami
4 corned beef
5 Bacon
6 pastrami
7 honey ham
8 nova lox
Name: food, dtype: object
# 在data中food列，有些存在大写，所以需要处理一下。
In [50]: data['food'].map(str.lower).map(meat_to_animal)
Out[50]:
0 pig
1 pig
2 pig
3 cow
4 cow
5 pig
6 cow
7 pig
8 salmon
Name: food, dtype: object
# 通过字典进行映射
In [51]: data['animal']=data['food'].map(str.lower).map(meat_to_animal)
In [52]: data
Out[52]:
food ounces animal
0 bacon 4.0 pig
1 pulled pork 3.0 pig
2 bacon 12.0 pig
3 Pastrami 6.0 cow
4 corned beef 7.5 cow
5 Bacon 8.0 pig
6 pastrami 3.0 cow
7 honey ham 5.0 pig
8 nova lox 6.0 salmon
#通过函数进行映射
In [53]: data['animal']=data['food'].map(lambda x:meat_to_animal[(x.lower())])
In [54]: data
Out[54]:
food ounces animal
0 bacon 4.0 pig
1 pulled pork 3.0 pig
2 bacon 12.0 pig
3 Pastrami 6.0 cow
4 corned beef 7.5 cow
5 Bacon 8.0 pig
6 pastrami 3.0 cow
7 honey ham 5.0 pig
8 nova lox 6.0 salmon
3.3 替换值
fillna是填充nan值，可以看成是替换值的一种特殊情况。
replace方法提供一个简单的替换值得方法。
In [55]: data = Series([1,-999,2,-999,-1000,3])
In [56]: data
Out[56]:
0 1
1 -999
2 2
3 -999
4 -1000
5 3
dtype: int64
In [57]: data.replace(-999,np.nan)
Out[57]:
0 1.0
1 NaN
2 2.0
3 NaN
4 -1000.0
5 3.0
dtype: float64
In [58]: data.replace([-999,-1000],np.nan) # 列表 + 标量
Out[58]:
0 1.0
1 NaN
2 2.0
3 NaN
4 NaN
5 3.0
dtype: float64
In [59]: data.replace([-999,-1000],[np.nan,0]) # 两个列表
Out[59]:
0 1.0
1 NaN
2 2.0
3 NaN
4 0.0
5 3.0
dtype: float64
In [60]: data.replace({-999:np.nan,-1000:0}) # 字典形式
Out[60]:
0 1.0
1 NaN
2 2.0
3 NaN
4 0.0
5 3.0
dtype: float64
3.4 重命名轴索引
# 修改源数据
In [62]: data
Out[62]:
0 1
1 -999
2 2
3 -999
4 -1000
5 3
dtype: int64
In [63]: data.index = range(2,8)
In [64]: data
Out[64]:
2 1
3 -999
4 2
5 -999
6 -1000
7 3
dtype: int64
In [65]: data.index.map(lambda x:x+100)
Out[65]: array([102, 103, 104, 105, 106, 107])
In [67]: data.index = data.index.map(lambda x:x+100)
In [68]: data
Out[68]:
102 1
103 -999
104 2
105 -999
106 -1000
107 3
dtype: int64
# 创建一个新数据，使用rename
In [73]: data2 = DataFrame(np.arange(12).reshape(3,4),columns=list('abcd'),index
...: =range(3))
In [74]: data2
Out[74]:
a b c d
0 0 1 2 3
1 4 5 6 7
2 8 9 10 11
In [77]: data2.rename(index={0:100})
Out[77]:
a b c d
100 0 1 2 3
1 4 5 6 7
2 8 9 10 11
In [78]: data2.rename(index={0:100},columns={'a':'ff'})
Out[78]:
ff b c d
100 0 1 2 3
1 4 5 6 7
2 8 9 10 11
3.5 离散化和面元划分
连续数据常常需要被拆分为离散数据。这就需要用cut函数。
In [79]: ages = range(100)
In [80]: bins = [18,25,35,60,100]
In [81]: pd.cut(ages,bins)
Out[81]:
[NaN, NaN, NaN, NaN, NaN, ..., (60, 100], (60, 100], (60, 100], (60, 100], (60, 100]]
Length: 100
Categories (4, object): [(18, 25] < (25, 35] < (35, 60] < (60, 100]]
In [82]: cuts = pd.cut(ages,bins)
In [83]: cuts[45]
Out[83]: '(35, 60]'
In [85]: cuts.dtype
Out[85]: category
这时，cut返回的是一个category对象，
In [86]: cuts.labels
/Users/yangfeilong/anaconda/bin/ipython:1: FutureWarning: 'labels' is deprecated. Use 'codes' instead
#!/bin/bash /Users/yangfeilong/anaconda/bin/python.app
Out[86]:
array([-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
-1, -1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3], dtype=int8)
In [87]: cuts.levels
/Users/yangfeilong/anaconda/bin/ipython:1: FutureWarning: Accessing 'levels' is deprecated, use 'categories'
#!/bin/bash /Users/yangfeilong/anaconda/bin/python.app
Out[87]: Index([u'(18, 25]', u'(25, 35]', u'(35, 60]', u'(60, 100]'], dtype='object')
In [88]: pd.value_counts(cuts) #查看每个种类的数量
Out[88]:
(60, 100] 39
(35, 60] 25
(25, 35] 10
(18, 25] 7
dtype: int64
# 使用right=False可以修改开端和闭端
In [89]: cuts = pd.cut(ages,bins,right=False)
In [90]: cuts
Out[90]:
[NaN, NaN, NaN, NaN, NaN, ..., [60, 100), [60, 100), [60, 100), [60, 100), [60, 100)]
Length: 100
Categories (4, object): [[18, 25) < [25, 35) < [35, 60) < [60, 100)]
In [91]: cuts = pd.cut(ages,bins,right=False,labels=list('abcd'))
In [92]: cuts
Out[92]:
[NaN, NaN, NaN, NaN, NaN, ..., d, d, d, d, d]
Length: 100
Categories (4, object): [a < b < c < d]
# 如果cut传入的是数字n，那么就会均分成n份。
In [94]: pd.cut(ages,5)
Out[94]:
[(-0.099, 19.8], (-0.099, 19.8], (-0.099, 19.8], (-0.099, 19.8], (-0.099, 19.8], ..., (79.2, 99], (79.2, 99], (79.2, 99], (79.2, 99], (79.2, 99]]
Length: 100
Categories (5, object): [(-0.099, 19.8] < (19.8, 39.6] < (39.6, 59.4] < (59.4, 79.2] < (79.2, 99]]
# qcut使用样本分位数进行划分，比cut更平均，获得大小基本相同的面元
In [95]: pd.qcut(ages,5)
Out[95]:
[[0, 19.8], [0, 19.8], [0, 19.8], [0, 19.8], [0, 19.8], ..., (79.2, 99], (79.2, 99], (79.2, 99], (79.2, 99], (79.2, 99]]
Length: 100
Categories (5, object): [[0, 19.8] < (19.8, 39.6] < (39.6, 59.4] < (59.4, 79.2] < (79.2, 99]]
In [96]: pd.value_counts(pd.qcut(ages,5))
Out[96]:
(79.2, 99] 20
(59.4, 79.2] 20
(39.6, 59.4] 20
(19.8, 39.6] 20
[0, 19.8] 20
dtype: int64
In [98]: ages = np.random.randn(1000)
In [99]: pd.value_counts(pd.qcut(ages,5))
Out[99]:
(0.883, 3.534] 200
(0.25, 0.883] 200
(-0.266, 0.25] 200
(-0.806, -0.266] 200
[-3.172, -0.806] 200
dtype: int64
In [100]: pd.value_counts(pd.cut(ages,5))
Out[100]:
(-0.489, 0.852] 481
(-1.831, -0.489] 283
(0.852, 2.193] 198
(-3.179, -1.831] 23
(2.193, 3.534] 15
dtype: int64
3.6 检测和过滤异常值
只看个例子：
In [101]: data = DataFrame(np.random.randn(1000,4))
In [102]: data
Out[102]:
0 1 2 3
0 0.746898 -0.657217 0.358517 1.075503
1 -0.821582 1.408816 0.937318 -0.548889
2 -0.527847 0.749256 -0.524283 -0.977338
3 0.800118 -0.544609 0.481839 0.555546
...
996 -0.081920 0.100507 -0.431551 -1.832261
997 -1.069475 0.205978 0.918592 0.881498
998 1.749007 2.658384 -1.079657 -0.023890
999 1.488167 -0.061146 0.006189 0.070564
[1000 rows x 4 columns]
In [103]: data.describe()
Out[103]:
0 1 2 3
count 1000.000000 1000.000000 1000.000000 1000.000000
mean -0.018474 0.030781 0.025966 0.041950
std 1.020634 0.982547 1.010164 0.948116
min -3.150278 -3.422772 -3.406069 -2.575391
25% -0.719646 -0.621461 -0.644949 -0.572558
50% -0.021287 0.043128 0.027890 0.050492
75% 0.701195 0.711455 0.678644 0.665413
max 3.270054 2.983500 3.323021 2.925822
In [105]: data[(np.abs(data) > 3).any(1)]
Out[105]:
0 1 2 3
16 1.450842 2.222636 -3.289273 1.265199
353 -3.067571 0.490152 0.536051 1.252143
531 1.518928 0.289636 3.323021 -0.695272
688 1.776682 -0.799736 -3.406069 0.632751
702 0.807953 -0.674675 -3.045511 -0.267613
704 -0.469205 -3.422772 -2.669146 -0.712407
772 -0.861015 -0.573432 3.012727 0.325476
808 3.270054 -0.035521 0.061092 1.263363
812 -3.150278 -0.734308 1.376565 0.671424
In [108]: data[np.abs(data) > 3] = 100
In [110]: data.describe()
Out[110]:
0 1 2 3
count 1000.000000 1000.000000 1000.000000 1000.000000
mean 0.284474 0.134204 0.529371 0.041950
std 5.564261 3.308565 7.123116 0.948116
min -2.746446 -2.983108 -2.710984 -2.575391
25% -0.714890 -0.619551 -0.635457 -0.572558
50% -0.018859 0.048775 0.036321 0.050492
75% 0.710050 0.712375 0.686314 0.665413
max 100.000000 100.000000 100.000000 2.925822
In [111]: data[np.abs(data) > 3] = np.sign(data) * 3
In [113]: data.describe()
Out[113]:
0 1 2 3
count 1000.000000 1000.000000 1000.000000 1000.000000
mean -0.006526 0.037204 0.044371 0.041950
std 1.019295 0.980940 1.006138 0.948116
min -2.746446 -2.983108 -2.710984 -2.575391
25% -0.714890 -0.619551 -0.635457 -0.572558
50% -0.018859 0.048775 0.036321 0.050492
75% 0.710050 0.712375 0.686314 0.665413
max 3.000000 3.000000 3.000000 2.925822
3.7 排列和随机采样
# 随机重排序
In [114]: df = DataFrame(np.arange(20).reshape(5,4))
In [115]: df.ix[np.random.permutation(5)]
Out[115]:
0 1 2 3
0 0 1 2 3
4 16 17 18 19
1 4 5 6 7
2 8 9 10 11
3 12 13 14 15
# 获取随机前三行
In [116]: df.ix[np.random.permutation(len(df))[:3]]
Out[116]:
0 1 2 3
0 0 1 2 3
2 8 9 10 11
4 16 17 18 19
# 随机获取10，每个值随机取自bag
In [118]: bag = np.array([5,7,-1,4,6])
In [119]: bag.take(np.random.randint(0,len(bag),size=10))
Out[119]: array([ 7, 7, 7, 5, 7, 6, 4, -1, 6, 4])
3.8 计算标量/哑变量
还有一种常用的转换方式是：将分类变量转换为哑变量矩阵或指标矩阵。
# 比较复杂，其实就是用1和0填充一个矩阵，说明元素在一个Series中的位置。
In [120]: s1 = Series(list('bbacab'))
In [121]: pd.get_dummies(s1)
Out[121]:
a b c
0 0.0 1.0 0.0
1 0.0 1.0 0.0
2 1.0 0.0 0.0
3 0.0 0.0 1.0
4 1.0 0.0 0.0
5 0.0 1.0 0.0
结合cut和get_dummies函数可以来做一些统计有用的事情：
In [122]: values = np.random.rand(10)
In [123]: values
Out[123]:
array([ 0.96294606, 0.15142481, 0.38260636, 0.01174038, 0.44699216,
0.3726047 , 0.50965999, 0.61532174, 0.27740518, 0.28994142])
In [126]: bins = [0,0.2,0.4,0.6,0.8,1.]
In [127]: pd.cut(values,bins)
Out[127]:
[(0.8, 1], (0, 0.2], (0.2, 0.4], (0, 0.2], (0.4, 0.6], (0.2, 0.4], (0.4, 0.6], (0.6, 0.8], (0.2, 0.4], (0.2, 0.4]]
Categories (5, object): [(0, 0.2] < (0.2, 0.4] < (0.4, 0.6] < (0.6, 0.8] < (0.8, 1]]
In [128]: pd.get_dummies(pd.cut(values, bins)) # 查看每个类型的指标矩阵
Out[128]:
(0, 0.2] (0.2, 0.4] (0.4, 0.6] (0.6, 0.8] (0.8, 1]
0 0.0 0.0 0.0 0.0 1.0
1 1.0 0.0 0.0 0.0 0.0
2 0.0 1.0 0.0 0.0 0.0
3 1.0 0.0 0.0 0.0 0.0
4 0.0 0.0 1.0 0.0 0.0
5 0.0 1.0 0.0 0.0 0.0
6 0.0 0.0 1.0 0.0 0.0
7 0.0 0.0 0.0 1.0 0.0
8 0.0 1.0 0.0 0.0 0.0
9 0.0 1.0 0.0 0.0 0.0
In [129]: a =pd.get_dummies(pd.cut(values, bins))
In [131]: a.sum(axis=0) # 统计一下每个类型的数量
Out[131]:
(0, 0.2] 2.0
(0.2, 0.4] 4.0
(0.4, 0.6] 2.0
(0.6, 0.8] 1.0
(0.8, 1] 1.0
dtype: float64
待续。。。