十分钟搞定pandas

内容简介：十分钟搞定pandas

这是关于pandas的简短介绍，主要面向新用户。可以参阅 Cookbook 了解更复杂的使用方法。

习惯上,我们做以下导入

In [1]: import pandas as pd
In [2]: import numpy as np
In [3]: import matplotlib.pyplot as plt

创建对象

使用传递的值列表序列创建 序列 , 让pandas创建默认整数索引

In [4]: s = pd.Series([1,3,5,np.nan,6,8])
In [5]: s
Out[5]:
0     1
1     3
2     5
3   NaN
4     6
5     8
dtype: float64

使用传递的numpy数组创建 数据帧 ,并使用日期索引和标记列.

In [6]: dates = pd.date_range('20130101',periods=6)
In [7]: dates
Out[7]:
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-01, ..., 2013-01-06]
Length: 6, Freq: D, Timezone: None
 
In [8]: df = pd.DataFrame(np.random.randn(6,4),index=dates,columns=list('ABCD'))
In [9]: df
Out[9]:
                   A         B         C         D
2013-01-01  0.469112 -0.282863 -1.509059 -1.135632
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804
2013-01-04  0.721555 -0.706771 -1.039575  0.271860
2013-01-05 -0.424972  0.567020  0.276232 -1.087401
2013-01-06 -0.673690  0.113648 -1.478427  0.524988

使用传递的可转换序列的字典对象创建 数据帧 .

In [10]: df2 = pd.DataFrame({ 'A' : 1.,
   ....:                      'B' : pd.Timestamp('20130102'),
   ....:                      'C' : pd.Series(1,index=list(range(4)),dtype='float32'),
   ....:                      'D' : np.array([3] * 4,dtype='int32'),
   ....:                      'E' : pd.Categorical(["test","train","test","train"]),
   ....:                      'F' : 'foo' })
   ....:
In [11]: df2
Out[11]:
   A          B  C  D      E    F
0  1 2013-01-02  1  3   test  foo
1  1 2013-01-02  1  3  train  foo
2  1 2013-01-02  1  3   test  foo
3  1 2013-01-02  1  3  train  foo

所有明确 类型

In [12]: df2.dtypes
Out[12]:
A           float64
B    datetime64[ns]
C           float32
D             int32
E          category
F            object
dtype: object

如果你这个正在使用IPython，标签补全列名（以及公共属性）将自动启用。这里是将要完成的属性的子集：

In [13]: df2.<TAB>
df2.A                  df2.boxplot
df2.abs                df2.C
df2.add                df2.clip
df2.add_prefix         df2.clip_lower
df2.add_suffix         df2.clip_upper
df2.align              df2.columns
df2.all                df2.combine
df2.any                df2.combineAdd
df2.append             df2.combine_first
df2.apply              df2.combineMult
df2.applymap           df2.compound
df2.as_blocks          df2.consolidate
df2.asfreq             df2.convert_objects
df2.as_matrix          df2.copy
df2.astype             df2.corr
df2.at                 df2.corrwith
df2.at_time            df2.count
df2.axes               df2.cov
df2.B                  df2.cummax
df2.between_time       df2.cummin
df2.bfill              df2.cumprod
df2.blocks             df2.cumsum
df2.bool               df2.D

如你所见, 列 A , B , C , 和 D 也是自动完成标签. E 也是可用的; 为了简便起见,后面的属性显示被截断.

查看数据

参阅 基础部分

查看帧顶部和底部行

In [14]: df.head()
Out[14]:
                   A         B         C         D
2013-01-01  0.469112 -0.282863 -1.509059 -1.135632
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804
2013-01-04  0.721555 -0.706771 -1.039575  0.271860
2013-01-05 -0.424972  0.567020  0.276232 -1.087401
 
In [15]: df.tail(3)
Out[15]:
                   A         B         C         D
2013-01-04  0.721555 -0.706771 -1.039575  0.271860
2013-01-05 -0.424972  0.567020  0.276232 -1.087401
2013-01-06 -0.673690  0.113648 -1.478427  0.524988

显示索引,列,和底层numpy数据

In [16]: df.index
Out[16]:
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-01, ..., 2013-01-06]
Length: 6, Freq: D, Timezone: None
 
In [17]: df.columns
Out[17]: Index([u'A', u'B', u'C', u'D'], dtype='object')
 
In [18]: df.values
Out[18]:
array([[ 0.4691, -0.2829, -1.5091, -1.1356],
       [ 1.2121, -0.1732,  0.1192, -1.0442],
       [-0.8618, -2.1046, -0.4949,  1.0718],
       [ 0.7216, -0.7068, -1.0396,  0.2719],
       [-0.425 ,  0.567 ,  0.2762, -1.0874],
       [-0.6737,  0.1136, -1.4784,  0.525 ]])

描述显示数据快速统计摘要

In [19]: df.describe()
Out[19]:
              A         B         C         D
count  6.000000  6.000000  6.000000  6.000000
mean   0.073711 -0.431125 -0.687758 -0.233103
std    0.843157  0.922818  0.779887  0.973118
min   -0.861849 -2.104569 -1.509059 -1.135632
25%   -0.611510 -0.600794 -1.368714 -1.076610
50%    0.022070 -0.228039 -0.767252 -0.386188
75%    0.658444  0.041933 -0.034326  0.461706
max    1.212112  0.567020  0.276232  1.071804

转置数据

In [20]: df.T
Out[20]:
   2013-01-01  2013-01-02  2013-01-03  2013-01-04  2013-01-05  2013-01-06
A    0.469112    1.212112   -0.861849    0.721555   -0.424972   -0.673690
B   -0.282863   -0.173215   -2.104569   -0.706771    0.567020    0.113648
C   -1.509059    0.119209   -0.494929   -1.039575    0.276232   -1.478427
D   -1.135632   -1.044236    1.071804    0.271860   -1.087401    0.524988

按轴排序

In [21]: df.sort_index(axis=1, ascending=False)
Out[21]:
                   D         C         B         A
2013-01-01 -1.135632 -1.509059 -0.282863  0.469112
2013-01-02 -1.044236  0.119209 -0.173215  1.212112
2013-01-03  1.071804 -0.494929 -2.104569 -0.861849
2013-01-04  0.271860 -1.039575 -0.706771  0.721555
2013-01-05 -1.087401  0.276232  0.567020 -0.424972
2013-01-06  0.524988 -1.478427  0.113648 -0.673690

按值排序

In [22]: df.sort(columns='B')
Out[22]:
                   A         B         C         D
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804
2013-01-04  0.721555 -0.706771 -1.039575  0.271860
2013-01-01  0.469112 -0.282863 -1.509059 -1.135632
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-06 -0.673690  0.113648 -1.478427  0.524988
2013-01-05 -0.424972  0.567020  0.276232 -1.087401

选择器

注释:标准Python / Numpy表达式可以完成这些互动工作, 但在生产代码中, 我们推荐使用优化的pandas数据访问方法, .at, .iat, .loc, .iloc 和 .ix.

参阅索引文档 索引和选择数据 and 多索引/高级索引

读取

选择单列, 这会产生一个 序列 , 等价df. A

In [23]: df['A']
Out[23]:
2013-01-01    0.469112
2013-01-02    1.212112
2013-01-03   -0.861849
2013-01-04    0.721555
2013-01-05   -0.424972
2013-01-06   -0.673690
Freq: D, Name: A, dtype: float64

使用 [] 选择行片断

In [24]: df[0:3]
Out[24]:
                   A         B         C         D
2013-01-01  0.469112 -0.282863 -1.509059 -1.135632
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804
 
In [25]: df['20130102':'20130104']
Out[25]:
                   A         B         C         D
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804
2013-01-04  0.721555 -0.706771 -1.039575  0.271860

使用标签选择

更多信息请参阅 按标签选择

使用标签获取横截面

In [26]: df.loc[dates[0]]
Out[26]:
A    0.469112
B   -0.282863
C   -1.509059
D   -1.135632
Name: 2013-01-01 00:00:00, dtype: float64

使用标签选择多轴

In [27]: df.loc[:,['A','B']]
Out[27]:
                   A         B
2013-01-01  0.469112 -0.282863
2013-01-02  1.212112 -0.173215
2013-01-03 -0.861849 -2.104569
2013-01-04  0.721555 -0.706771
2013-01-05 -0.424972  0.567020
2013-01-06 -0.673690  0.113648

显示标签切片, 包含 两个端点

In [28]: df.loc['20130102':'20130104',['A','B']]
Out[28]:
                   A         B
2013-01-02  1.212112 -0.173215
2013-01-03 -0.861849 -2.104569
2013-01-04  0.721555 -0.706771

降低返回对象维度

In [29]: df.loc['20130102',['A','B']]
Out[29]:
A    1.212112
B   -0.173215
Name: 2013-01-02 00:00:00, dtype: float64

获取标量值

In [30]: df.loc[dates[0],'A']
Out[30]: 0.46911229990718628

快速访问并获取标量数据 (等价上面的方法)

In [31]: df.at[dates[0],'A']
Out[31]: 0.46911229990718628

按位置选择

更多信息请参阅 按位置参阅

传递整数选择位置

In [32]: df.iloc[3]
Out[32]:
A    0.721555
B   -0.706771
C   -1.039575
D    0.271860
Name: 2013-01-04 00:00:00, dtype: float64

使用整数片断,效果类似numpy/python

In [33]: df.iloc[3:5,0:2]
Out[33]:
                   A         B
2013-01-04  0.721555 -0.706771
2013-01-05 -0.424972  0.567020

使用整数偏移定位列表,效果类似 numpy/python 样式

In [34]: df.iloc[[1,2,4],[0,2]]
Out[34]:
                   A         C
2013-01-02  1.212112  0.119209
2013-01-03 -0.861849 -0.494929
2013-01-05 -0.424972  0.276232

显式行切片

In [35]: df.iloc[1:3,:]
Out[35]:
                   A         B         C         D
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804

显式列切片

In [36]: df.iloc[:,1:3]
Out[36]:
                   B         C
2013-01-01 -0.282863 -1.509059
2013-01-02 -0.173215  0.119209
2013-01-03 -2.104569 -0.494929
2013-01-04 -0.706771 -1.039575
2013-01-05  0.567020  0.276232
2013-01-06  0.113648 -1.478427

显式获取一个值

In [37]: df.iloc[1,1]
Out[37]: -0.17321464905330861

快速访问一个标量（等同上个方法）

In [38]: df.iat[1,1]
Out[38]: -0.17321464905330861

布尔索引

使用单个列的值选择数据.

In [39]: df[df.A > 0]
Out[39]:
                   A         B         C         D
2013-01-01  0.469112 -0.282863 -1.509059 -1.135632
2013-01-02  1.212112 -0.173215  0.119209 -1.044236
2013-01-04  0.721555 -0.706771 -1.039575  0.271860

where 操作.

In [40]: df[df > 0]
Out[40]:
                   A         B         C         D
2013-01-01  0.469112       NaN       NaN       NaN
2013-01-02  1.212112       NaN  0.119209       NaN
2013-01-03       NaN       NaN       NaN  1.071804
2013-01-04  0.721555       NaN       NaN  0.271860
2013-01-05       NaN  0.567020  0.276232       NaN
2013-01-06       NaN  0.113648       NaN  0.524988

使用 isin() 筛选：

In [41]: df2 = df.copy()
In [42]: df2['E']=['one', 'one','two','three','four','three']
 
In [43]: df2
Out[43]:
                   A         B         C         D      E
2013-01-01  0.469112 -0.282863 -1.509059 -1.135632    one
2013-01-02  1.212112 -0.173215  0.119209 -1.044236    one
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804    two
2013-01-04  0.721555 -0.706771 -1.039575  0.271860  three
2013-01-05 -0.424972  0.567020  0.276232 -1.087401   four
2013-01-06 -0.673690  0.113648 -1.478427  0.524988  three
 
In [44]: df2[df2['E'].isin(['two','four'])]
Out[44]:
                   A         B         C         D     E
2013-01-03 -0.861849 -2.104569 -0.494929  1.071804   two
2013-01-05 -0.424972  0.567020  0.276232 -1.087401  four

赋值

赋值一个新列，通过索引自动对齐数据

In [45]: s1 = pd.Series([1,2,3,4,5,6],index=pd.date_range('20130102',periods=6))
In [46]: s1
Out[46]:
2013-01-02    1
2013-01-03    2
2013-01-04    3
2013-01-05    4
2013-01-06    5
2013-01-07    6
Freq: D, dtype: int64
 
In [47]: df['F'] = s1

按标签赋值

In [48]: df.at[dates[0],'A'] = 0

按位置赋值

In [49]: df.iat[0,1] = 0

通过numpy数组分配赋值

In [50]: df.loc[:,'D'] = np.array([5] * len(df))

之前的操作结果

In [51]: df
Out[51]:
                   A         B         C  D   F
2013-01-01  0.000000  0.000000 -1.509059  5 NaN
2013-01-02  1.212112 -0.173215  0.119209  5   1
2013-01-03 -0.861849 -2.104569 -0.494929  5   2
2013-01-04  0.721555 -0.706771 -1.039575  5   3
2013-01-05 -0.424972  0.567020  0.276232  5   4
2013-01-06 -0.673690  0.113648 -1.478427  5   5

where操作赋值.

In [52]: df2 = df.copy()
In [53]: df2[df2 > 0] = -df2
In [54]: df2
Out[54]:
                   A         B         C  D   F
2013-01-01  0.000000  0.000000 -1.509059 -5 NaN
2013-01-02 -1.212112 -0.173215 -0.119209 -5  -1
2013-01-03 -0.861849 -2.104569 -0.494929 -5  -2
2013-01-04 -0.721555 -0.706771 -1.039575 -5  -3
2013-01-05 -0.424972 -0.567020 -0.276232 -5  -4
2013-01-06 -0.673690 -0.113648 -1.478427 -5  -5

丢失的数据

pandas主要使用 np.nan 替换丢失的数据. 默认情况下它并不包含在计算中. 请参阅 Missing Data section

重建索引允许更改/添加/删除指定轴索引,并返回数据副本.

In [55]: df1 = df.reindex(index=dates[0:4],columns=list(df.columns) + ['E'])
In [56]: df1.loc[dates[0]:dates[1],'E'] = 1
In [57]: df1
Out[57]:
                   A         B         C  D   F   E
2013-01-01  0.000000  0.000000 -1.509059  5 NaN   1
2013-01-02  1.212112 -0.173215  0.119209  5   1   1
2013-01-03 -0.861849 -2.104569 -0.494929  5   2 NaN
2013-01-04  0.721555 -0.706771 -1.039575  5   3 NaN

删除任何有丢失数据的行.

In [58]: df1.dropna(how='any')
Out[58]:
                   A         B         C  D  F  E
2013-01-02  1.212112 -0.173215  0.119209  5  1  1

填充丢失数据

In [59]: df1.fillna(value=5)
Out[59]:
                   A         B         C  D  F  E
2013-01-01  0.000000  0.000000 -1.509059  5  5  1
2013-01-02  1.212112 -0.173215  0.119209  5  1  1
2013-01-03 -0.861849 -2.104569 -0.494929  5  2  5
2013-01-04  0.721555 -0.706771 -1.039575  5  3  5

获取值是否 nan 的布尔标记

In [60]: pd.isnull(df1)
Out[60]:
                A      B      C      D      F      E
2013-01-01  False  False  False  False   True  False
2013-01-02  False  False  False  False  False  False
2013-01-03  False  False  False  False  False   True
2013-01-04  False  False  False  False  False   True

运算

参阅 二元运算基础

统计

计算时一般不包括丢失的数据

执行描述性统计

In [61]: df.mean()
Out[61]:
A   -0.004474
B   -0.383981
C   -0.687758
D    5.000000
F    3.000000
dtype: float64

在其他轴做相同的运算

In [62]: df.mean(1)
Out[62]:
2013-01-01    0.872735
2013-01-02    1.431621
2013-01-03    0.707731
2013-01-04    1.395042
2013-01-05    1.883656
2013-01-06    1.592306
Freq: D, dtype: float64

用于运算的对象有不同的维度并需要对齐.除此之外，pandas会自动沿着指定维度计算.

In [63]: s = pd.Series([1,3,5,np.nan,6,8],index=dates).shift(2)
In [64]: s
Out[64]:
2013-01-01   NaN
2013-01-02   NaN
2013-01-03     1
2013-01-04     3
2013-01-05     5
2013-01-06   NaN
Freq: D, dtype: float64
 
In [65]: df.sub(s,axis='index')
Out[65]:
                   A         B         C   D   F
2013-01-01       NaN       NaN       NaN NaN NaN
2013-01-02       NaN       NaN       NaN NaN NaN
2013-01-03 -1.861849 -3.104569 -1.494929   4   1
2013-01-04 -2.278445 -3.706771 -4.039575   2   0
2013-01-05 -5.424972 -4.432980 -4.723768   0  -1
2013-01-06       NaN       NaN       NaN NaN NaN

Apply

在数据上使用函数

In [66]: df.apply(np.cumsum)
Out[66]:
                   A         B         C   D   F
2013-01-01  0.000000  0.000000 -1.509059   5 NaN
2013-01-02  1.212112 -0.173215 -1.389850  10   1
2013-01-03  0.350263 -2.277784 -1.884779  15   3
2013-01-04  1.071818 -2.984555 -2.924354  20   6
2013-01-05  0.646846 -2.417535 -2.648122  25  10
2013-01-06 -0.026844 -2.303886 -4.126549  30  15
 
In [67]: df.apply(lambda x: x.max() - x.min())
Out[67]:
A    2.073961
B    2.671590
C    1.785291
D    0.000000
F    4.000000
dtype: float64

直方图

请参阅 直方图和离散化

In [68]: s = pd.Series(np.random.randint(0,7,size=10))
In [69]: s
Out[69]:
0    4
1    2
2    1
3    2
4    6
5    4
6    4
7    6
8    4
9    4
dtype: int32
 
In [70]: s.value_counts()
Out[70]:
4    5
6    2
2    2
1    1
dtype: int64

字符串方法

序列可以使用一些字符串处理方法很轻易操作数据组中的每个元素,比如以下代码片断。 注意字符匹配方法默认情况下通常使用 正则表达式 （并且大多数时候都如此）. 更多信息请参阅 字符串向量方法 .

In [71]: s = pd.Series(['A', 'B', 'C', 'Aaba', 'Baca', np.nan, 'CABA', 'dog', 'cat'])
In [72]: s.str.lower()
Out[72]:
0       a
1       b
2       c
3    aaba
4    baca
5     NaN
6    caba
7     dog
8     cat
dtype: object

合并

连接

pandas提供各种 工具 以简便合并序列,数据桢,和组合对象, 在连接/合并类型操作中使用多种类型索引和相关数学函数.

请参阅 合并部分

把pandas对象连接到一起

In [73]: df = pd.DataFrame(np.random.randn(10, 4))
In [74]: df
Out[74]:
          0         1         2         3
0 -0.548702  1.467327 -1.015962 -0.483075
1  1.637550 -1.217659 -0.291519 -1.745505
2 -0.263952  0.991460 -0.919069  0.266046
3 -0.709661  1.669052  1.037882 -1.705775
4 -0.919854 -0.042379  1.247642 -0.009920
5  0.290213  0.495767  0.362949  1.548106
6 -1.131345 -0.089329  0.337863 -0.945867
7 -0.932132  1.956030  0.017587 -0.016692
8 -0.575247  0.254161 -1.143704  0.215897
9  1.193555 -0.077118 -0.408530 -0.862495
 
# break it into pieces
In [75]: pieces = [df[:3], df[3:7], df[7:]]
In [76]: pd.concat(pieces)
Out[76]:
          0         1         2         3
0 -0.548702  1.467327 -1.015962 -0.483075
1  1.637550 -1.217659 -0.291519 -1.745505
2 -0.263952  0.991460 -0.919069  0.266046
3 -0.709661  1.669052  1.037882 -1.705775
4 -0.919854 -0.042379  1.247642 -0.009920
5  0.290213  0.495767  0.362949  1.548106
6 -1.131345 -0.089329  0.337863 -0.945867
7 -0.932132  1.956030  0.017587 -0.016692
8 -0.575247  0.254161 -1.143704  0.215897
9  1.193555 -0.077118 -0.408530 -0.862495

连接

SQL样式合并. 请参阅 数据库style联接

In [77]: left = pd.DataFrame({'key': ['foo', 'foo'], 'lval': [1, 2]})
In [78]: right = pd.DataFrame({'key': ['foo', 'foo'], 'rval': [4, 5]})
In [79]: left
Out[79]:
   key  lval
0  foo     1
1  foo     2
 
In [80]: right
Out[80]:
   key  rval
0  foo     4
1  foo     5
 
In [81]: pd.merge(left, right, on='key')
Out[81]:
   key  lval  rval
0  foo     1     4
1  foo     1     5
2  foo     2     4
3  foo     2     5

添加

添加行到数据增. 参阅 添加

In [82]: df = pd.DataFrame(np.random.randn(8, 4), columns=['A','B','C','D'])
In [83]: df
Out[83]:
          A         B         C         D
0  1.346061  1.511763  1.627081 -0.990582
1 -0.441652  1.211526  0.268520  0.024580
2 -1.577585  0.396823 -0.105381 -0.532532
3  1.453749  1.208843 -0.080952 -0.264610
4 -0.727965 -0.589346  0.339969 -0.693205
5 -0.339355  0.593616  0.884345  1.591431
6  0.141809  0.220390  0.435589  0.192451
7 -0.096701  0.803351  1.715071 -0.708758
 
In [84]: s = df.iloc[3]
In [85]: df.append(s, ignore_index=True)
Out[85]:
          A         B         C         D
0  1.346061  1.511763  1.627081 -0.990582
1 -0.441652  1.211526  0.268520  0.024580
2 -1.577585  0.396823 -0.105381 -0.532532
3  1.453749  1.208843 -0.080952 -0.264610
4 -0.727965 -0.589346  0.339969 -0.693205
5 -0.339355  0.593616  0.884345  1.591431
6  0.141809  0.220390  0.435589  0.192451
7 -0.096701  0.803351  1.715071 -0.708758
8  1.453749  1.208843 -0.080952 -0.264610

分组

对于“group by”指的是以下一个或多个处理

• 将数据按某些标准分割为不同的组
• 在每个独立组上应用函数
• 组合结果为一个数据结构

请参阅 分组部分

In [86]: 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)})
   ....:
In [87]: df
Out[87]:
     A      B         C         D
0  foo    one -1.202872 -0.055224
1  bar    one -1.814470  2.395985
2  foo    two  1.018601  1.552825
3  bar  three -0.595447  0.166599
4  foo    two  1.395433  0.047609
5  bar    two -0.392670 -0.136473
6  foo    one  0.007207 -0.561757
7  foo  three  1.928123 -1.623033

分组然后应用函数统计总和存放到结果组

In [88]: df.groupby('A').sum()
Out[88]:
            C        D
A                     
bar -2.802588  2.42611
foo  3.146492 -0.63958

按多列分组为层次索引,然后应用函数

In [89]: df.groupby(['A','B']).sum()
Out[89]:
                  C         D
A   B                        
bar one   -1.814470  2.395985
    three -0.595447  0.166599
    two   -0.392670 -0.136473
foo one   -1.195665 -0.616981
    three  1.928123 -1.623033
    two    2.414034  1.600434

重塑

请参阅章节 分层索引 和 重塑 .

堆叠

In [90]: tuples = list(zip(*[['bar', 'bar', 'baz', 'baz',
   ....:                      'foo', 'foo', 'qux', 'qux'],
   ....:                     ['one', 'two', 'one', 'two',
   ....:                      'one', 'two', 'one', 'two']]))
   ....:
In [91]: index = pd.MultiIndex.from_tuples(tuples, names=['first', 'second'])
In [92]: df = pd.DataFrame(np.random.randn(8, 2), index=index, columns=['A', 'B'])
In [93]: df2 = df[:4]
In [94]: df2
Out[94]:
                     A         B
first second                    
bar   one     0.029399 -0.542108
      two     0.282696 -0.087302
baz   one    -1.575170  1.771208
      two     0.816482  1.100230

堆叠函数 “压缩” 数据桢的列一个级别.

In [95]: stacked = df2.stack()
In [96]: stacked
Out[96]:
first  second   
bar    one     A    0.029399
               B   -0.542108
       two     A    0.282696
               B   -0.087302
baz    one     A   -1.575170
               B    1.771208
       two     A    0.816482
               B    1.100230
dtype: float64

被“堆叠”数据桢或序列(有 多个索引 作为 索引 ), 其 堆叠 的反向操作是 未堆栈 , 上面的数据默认反堆叠到 上一级别 :

In [97]: stacked.unstack()
Out[97]:
                     A         B
first second                    
bar   one     0.029399 -0.542108
      two     0.282696 -0.087302
baz   one    -1.575170  1.771208
      two     0.816482  1.100230
 
In [98]: stacked.unstack(1)
Out[98]:
second        one       two
first                      
bar   A  0.029399  0.282696
      B -0.542108 -0.087302
baz   A -1.575170  0.816482
      B  1.771208  1.100230
 
In [99]: stacked.unstack(0)
Out[99]:
first          bar       baz
second                      
one    A  0.029399 -1.575170
       B -0.542108  1.771208
two    A  0.282696  0.816482
       B -0.087302  1.100230

数据透视表

查看 数据透视表 .

In [100]: df = pd.DataFrame({'A' : ['one', 'one', 'two', 'three'] * 3,
   .....:                    'B' : ['A', 'B', 'C'] * 4,
   .....:                    'C' : ['foo', 'foo', 'foo', 'bar', 'bar', 'bar'] * 2,
   .....:                    'D' : np.random.randn(12),
   .....:                    'E' : np.random.randn(12)})
   .....:
In [101]: df
Out[101]:
        A  B    C         D         E
0     one  A  foo  1.418757 -0.179666
1     one  B  foo -1.879024  1.291836
2     two  C  foo  0.536826 -0.009614
3   three  A  bar  1.006160  0.392149
4     one  B  bar -0.029716  0.264599
5     one  C  bar -1.146178 -0.057409
6     two  A  foo  0.100900 -1.425638
7   three  B  foo -1.035018  1.024098
8     one  C  foo  0.314665 -0.106062
9     one  A  bar -0.773723  1.824375
10    two  B  bar -1.170653  0.595974
11  three  C  bar  0.648740  1.167115

我们可以从此数据非常容易的产生数据透视表:

In [102]: pd.pivot_table(df, values='D', index=['A', 'B'], columns=['C'])
Out[102]:
C             bar       foo
A     B                    
one   A -0.773723  1.418757
      B -0.029716 -1.879024
      C -1.146178  0.314665
three A  1.006160       NaN
      B       NaN -1.035018
      C  0.648740       NaN
two   A       NaN  0.100900
      B -1.170653       NaN
      C       NaN  0.536826

时间序列

pandas有易用,强大且高效的函数用于高频数据重采样转换操作(例如,转换秒数据到5分钟数据), 这是很普遍的情况，但并不局限于金融应用, 请参阅 时间序列章节

In [103]: rng = pd.date_range('1/1/2012', periods=100, freq='S')
In [104]: ts = pd.Series(np.random.randint(0, 500, len(rng)), index=rng)
In [105]: ts.resample('5Min', how='sum')
Out[105]:
2012-01-01    25083
Freq: 5T, dtype: int32

时区表示

In [106]: rng = pd.date_range('3/6/2012 00:00', periods=5, freq='D')
In [107]: ts = pd.Series(np.random.randn(len(rng)), rng)
In [108]: ts
Out[108]:
2012-03-06    0.464000
2012-03-07    0.227371
2012-03-08   -0.496922
2012-03-09    0.306389
2012-03-10   -2.290613
Freq: D, dtype: float64
 
In [109]: ts_utc = ts.tz_localize('UTC')
In [110]: ts_utc
Out[110]:
2012-03-06 00:00:00+00:00    0.464000
2012-03-07 00:00:00+00:00    0.227371
2012-03-08 00:00:00+00:00   -0.496922
2012-03-09 00:00:00+00:00    0.306389
2012-03-10 00:00:00+00:00   -2.290613
Freq: D, dtype: float64

转换到其它时区

In [111]: ts_utc.tz_convert('US/Eastern')
Out[111]:
2012-03-05 19:00:00-05:00    0.464000
2012-03-06 19:00:00-05:00    0.227371
2012-03-07 19:00:00-05:00   -0.496922
2012-03-08 19:00:00-05:00    0.306389
2012-03-09 19:00:00-05:00   -2.290613
Freq: D, dtype: float64

转换不同的时间跨度

In [112]: rng = pd.date_range('1/1/2012', periods=5, freq='M')
In [113]: ts = pd.Series(np.random.randn(len(rng)), index=rng)
In [114]: ts
Out[114]:
2012-01-31   -1.134623
2012-02-29   -1.561819
2012-03-31   -0.260838
2012-04-30    0.281957
2012-05-31    1.523962
Freq: M, dtype: float64
 
In [115]: ps = ts.to_period()
In [116]: ps
Out[116]:
2012-01   -1.134623
2012-02   -1.561819
2012-03   -0.260838
2012-04    0.281957
2012-05    1.523962
Freq: M, dtype: float64
 
In [117]: ps.to_timestamp()
Out[117]:
2012-01-01   -1.134623
2012-02-01   -1.561819
2012-03-01   -0.260838
2012-04-01    0.281957
2012-05-01    1.523962
Freq: MS, dtype: float64

转换时段并且使用一些运算函数, 下例中, 我们转换年报11月到季度结束每日上午9点数据

In [118]: prng = pd.period_range('1990Q1', '2000Q4', freq='Q-NOV')
In [119]: ts = pd.Series(np.random.randn(len(prng)), prng)
In [120]: ts.index = (prng.asfreq('M', 'e') + 1).asfreq('H', 's') + 9
In [121]: ts.head()
Out[121]:
1990-03-01 09:00   -0.902937
1990-06-01 09:00    0.068159
1990-09-01 09:00   -0.057873
1990-12-01 09:00   -0.368204
1991-03-01 09:00   -1.144073
Freq: H, dtype: float64

分类

自版本0.15起, pandas可以在 数据桢 中包含分类. 完整的文档, 请查看 分类介绍 and the API文档 .

In [122]: df = pd.DataFrame({"id":[1,2,3,4,5,6], "raw_grade":['a', 'b', 'b', 'a', 'a', 'e']})

转换原始类别为分类数据类型.

In [123]: df["grade"] = df["raw_grade"].astype("category")
In [124]: df["grade"]
Out[124]:
0    a
1    b
2    b
3    a
4    a
5    e
Name: grade, dtype: category
Categories (3, object): [a, b, e]

重命令分类为更有意义的名称 (分配到 Series.cat.categories 对应位置!)

In [125]: df["grade"].cat.categories = ["very good", "good", "very bad"]

重排顺分类,同时添加缺少的分类( 序列 .cat 方法下返回新默认 序列 )

In [126]: df["grade"] = df["grade"].cat.set_categories(["very bad", "bad", "medium", "good", "very good"])
In [127]: df["grade"]
Out[127]:
0    very good
1         good
2         good
3    very good
4    very good
5     very bad
Name: grade, dtype: category
Categories (5, object): [very bad, bad, medium, good, very good]

排列分类中的顺序,不是按词汇排列.

In [128]: df.sort("grade")
Out[128]:
   id raw_grade      grade
5   6         e   very bad
1   2         b       good
2   3         b       good
0   1         a  very good
3   4         a  very good
4   5         a  very good

类别列分组,并且也显示空类别.

In [129]: df.groupby("grade").size()
Out[129]:
grade
very bad      1
bad         NaN
medium      NaN
good          2
very good     3
dtype: float64

绘图

绘图 文档.

In [130]: ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2000', periods=1000))
In [131]: ts = ts.cumsum()
In [132]: ts.plot()
Out[132]: <matplotlib.axes._subplots.AxesSubplot at 0xb02091ac>

在数据桢中,可以很方便的绘制带标签列:

In [133]: df = pd.DataFrame(np.random.randn(1000, 4), index=ts.index,
   .....:                   columns=['A', 'B', 'C', 'D'])
   .....:
In [134]: df = df.cumsum()
In [135]: plt.figure(); df.plot(); plt.legend(loc='best')
Out[135]: <matplotlib.legend.Legend at 0xb01c9cac>

获取数据输入/输出

CSV

写入csv文件

In [136]: df.to_csv('foo.csv')

读取csv文件

In [137]: pd.read_csv('foo.csv')
Out[137]:
     Unnamed: 0          A          B         C          D
0    2000-01-01   0.266457  -0.399641 -0.219582   1.186860
1    2000-01-02  -1.170732  -0.345873  1.653061  -0.282953
2    2000-01-03  -1.734933   0.530468  2.060811  -0.515536
3    2000-01-04  -1.555121   1.452620  0.239859  -1.156896
4    2000-01-05   0.578117   0.511371  0.103552  -2.428202
5    2000-01-06   0.478344   0.449933 -0.741620  -1.962409
6    2000-01-07   1.235339  -0.091757 -1.543861  -1.084753
..          ...        ...        ...       ...        ...
993  2002-09-20 -10.628548  -9.153563 -7.883146  28.313940
994  2002-09-21 -10.390377  -8.727491 -6.399645  30.914107
995  2002-09-22  -8.985362  -8.485624 -4.669462  31.367740
996  2002-09-23  -9.558560  -8.781216 -4.499815  30.518439
997  2002-09-24  -9.902058  -9.340490 -4.386639  30.105593
998  2002-09-25 -10.216020  -9.480682 -3.933802  29.758560
999  2002-09-26 -11.856774 -10.671012 -3.216025  29.369368
 
[1000 rows x 5 columns]

HDF5

读写 HDF存储

写入HDF5存储

In [138]: df.to_hdf('foo.h5','df')

读取HDF5存储

In [139]: pd.read_hdf('foo.h5','df')
Out[139]:
                    A          B         C          D
2000-01-01   0.266457  -0.399641 -0.219582   1.186860
2000-01-02  -1.170732  -0.345873  1.653061  -0.282953
2000-01-03  -1.734933   0.530468  2.060811  -0.515536
2000-01-04  -1.555121   1.452620  0.239859  -1.156896
2000-01-05   0.578117   0.511371  0.103552  -2.428202
2000-01-06   0.478344   0.449933 -0.741620  -1.962409
2000-01-07   1.235339  -0.091757 -1.543861  -1.084753
...               ...        ...       ...        ...
2002-09-20 -10.628548  -9.153563 -7.883146  28.313940
2002-09-21 -10.390377  -8.727491 -6.399645  30.914107
2002-09-22  -8.985362  -8.485624 -4.669462  31.367740
2002-09-23  -9.558560  -8.781216 -4.499815  30.518439
2002-09-24  -9.902058  -9.340490 -4.386639  30.105593
2002-09-25 -10.216020  -9.480682 -3.933802  29.758560
2002-09-26 -11.856774 -10.671012 -3.216025  29.369368
 
[1000 rows x 4 columns]

Excel

读写 MS Excel

写入excel文件

In [140]: df.to_excel('foo.xlsx', sheet_name='Sheet1')

读取excel文件

In [141]: pd.read_excel('foo.xlsx', 'Sheet1', index_col=None, na_values=['NA'])
Out[141]:
                    A          B         C          D
2000-01-01   0.266457  -0.399641 -0.219582   1.186860
2000-01-02  -1.170732  -0.345873  1.653061  -0.282953
2000-01-03  -1.734933   0.530468  2.060811  -0.515536
2000-01-04  -1.555121   1.452620  0.239859  -1.156896
2000-01-05   0.578117   0.511371  0.103552  -2.428202
2000-01-06   0.478344   0.449933 -0.741620  -1.962409
2000-01-07   1.235339  -0.091757 -1.543861  -1.084753
...               ...        ...       ...        ...
2002-09-20 -10.628548  -9.153563 -7.883146  28.313940
2002-09-21 -10.390377  -8.727491 -6.399645  30.914107
2002-09-22  -8.985362  -8.485624 -4.669462  31.367740
2002-09-23  -9.558560  -8.781216 -4.499815  30.518439
2002-09-24  -9.902058  -9.340490 -4.386639  30.105593
2002-09-25 -10.216020  -9.480682 -3.933802  29.758560
2002-09-26 -11.856774 -10.671012 -3.216025  29.369368
 
[1000 rows x 4 columns]

陷阱

如果尝试这样操作可能会看到像这样的异常:

>>> if pd.Series([False, True, False]):
    print("I was true")
Traceback
    ...
ValueError: The truth value of an array is ambiguous. Use a.empty, a.any() or a.all().

查看 对照 获取解释和怎么做的帮助

也可以查看 陷阱 .

以上就是本文的全部内容，希望对大家的学习有所帮助，也希望大家多多支持 码农网

查看所有标签

猜你喜欢:

• 轻松搞定RocketMQ入门
• 索引，一文搞定
• 一次性搞定 Iconfont
• 简单搞定Nginx日志分割
• 十分钟搞定 Gradle
• 十分钟搞定 Gradle

本站部分资源来源于网络，本站转载出于传递更多信息之目的，版权归原作者或者来源机构所有，如转载稿涉及版权问题，请联系我们。