Pandas DataFrame 数据选取和过滤
Sat 01 April 2017
Pandas 是 Python Data Analysis Library, 是基于 numpy 库的一个为了数据分析而设计的一个 Python 库。它提供了很多工具和方法，使得使用 python 操作大量的数据变得高效而方便。
本文专门介绍 Pandas 中对 DataFrame 的一些对数据进行过滤、选取的方法和工具。 首先，本文所用的原始数据如下：
df = pd.DataFrame(np.random.randn(6,4), index=dates, columns=list('ABCD'))
Out[9]:
A B C D
2017-04-01 0.522241 0.495106 -0.268194 -0.035003
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926
2017-04-03 0.480507 1.215048 1.313314 -0.072320
2017-04-04 1.700309 0.287588 -0.012103 0.525291
2017-04-05 0.526615 -0.417645 0.405853 -0.835213
2017-04-06 1.143858 -0.326720 1.425379 0.531037
选取
通过 [] 来选取
选取一列或者几列：
df['A']
Out:
2017-04-01 0.522241
2017-04-02 2.104572
2017-04-03 0.480507
2017-04-04 1.700309
2017-04-05 0.526615
2017-04-06 1.143858
df[['A','B']]
Out:
A B
2017-04-01 0.522241 0.495106
2017-04-02 2.104572 -0.977768
2017-04-03 0.480507 1.215048
2017-04-04 1.700309 0.287588
2017-04-05 0.526615 -0.417645
2017-04-06 1.143858 -0.326720
选取某一行或者几行：
df['2017-04-01':'2017-04-01']
Out:
A B C D
2017-04-01 0.522241 0.495106 -0.268194 -0.03500
df['2017-04-01':'2017-04-03']
A B C D
2017-04-01 0.522241 0.495106 -0.268194 -0.035003
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926
2017-04-03 0.480507 1.215048 1.313314 -0.072320
loc, 通过行标签选取数据
df.loc['2017-04-01','A']
df.loc['2017-04-01']
Out:
A 0.522241
B 0.495106
C -0.268194
D -0.035003
df.loc['2017-04-01':'2017-04-03']
Out:
A B C D
2017-04-01 0.522241 0.495106 -0.268194 -0.035003
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926
2017-04-03 0.480507 1.215048 1.313314 -0.072320
df.loc['2017-04-01':'2017-04-04',['A','B']]
Out:
A B
2017-04-01 0.522241 0.495106
2017-04-02 2.104572 -0.977768
2017-04-03 0.480507 1.215048
2017-04-04 1.700309 0.287588
df.loc[:,['A','B']]
Out:
A B
2017-04-01 0.522241 0.495106
2017-04-02 2.104572 -0.977768
2017-04-03 0.480507 1.215048
2017-04-04 1.700309 0.287588
2017-04-05 0.526615 -0.417645
2017-04-06 1.143858 -0.326720
iloc, 通过行号获取数据
df.iloc[2]
Out:
A 0.480507
B 1.215048
C 1.313314
D -0.072320
df.iloc[1:3]
Out:
A B C D
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926
2017-04-03 0.480507 1.215048 1.313314 -0.072320
df.iloc[1,1]
df.iloc[1:3,1]
df.iloc[1:3,1:2]
df.iloc[[1,3],[2,3]]
Out:
C D
2017-04-02 -0.139632 -0.735926
2017-04-04 -0.012103 0.525291
df.iloc[[1,3],:]
df.iloc[:,[2,3]]
iat, 获取某一个 cell 的值
df.iat[1,2]
Out:
-0.13963224781812655
过滤
使用 [] 过滤
[]中是一个boolean 表达式，凡是计算为 True 的行就会被选取。
df[df.A>1]
Out:
A B C D
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926
2017-04-04 1.700309 0.287588 -0.012103 0.525291
2017-04-06 1.143858 -0.326720 1.425379 0.531037
df[df>1]
Out:
A B C D
2017-04-01 NaN NaN NaN NaN
2017-04-02 2.104572 NaN NaN NaN
2017-04-03 NaN 1.215048 1.313314 NaN
2017-04-04 1.700309 NaN NaN NaN
2017-04-05 NaN NaN NaN NaN
2017-04-06 1.143858 NaN 1.425379 NaN
df[df.A+df.B>1.5]
Out:
A B C D
2017-04-03 0.480507 1.215048 1.313314 -0.072320
2017-04-04 1.700309 0.287588 -0.012103 0.525291
下面是一个更加复杂的例子，选取的是 index 在 '2017-04-01'中'2017-04-04'的，一行的数据的和大于1的行：
df.loc['2017-04-01':'2017-04-04',df.sum()>1]
还可以通过和 apply 方法结合，构造更加复杂的过滤，实现将某个返回值为 boolean 的方法作为过滤条件：
df[df.apply(lambda x: x['b'] > x['c'], axis=1)]
使用 isin
df['E']=['one', 'one','two','three','four','three']
A B C D E
2017-04-01 0.522241 0.495106 -0.268194 -0.035003 one
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926 one
2017-04-03 0.480507 1.215048 1.313314 -0.072320 two
2017-04-04 1.700309 0.287588 -0.012103 0.525291 three
2017-04-05 0.526615 -0.417645 0.405853 -0.835213 four
2017-04-06 1.143858 -0.326720 1.425379 0.531037 three
df[df.E.isin(['one'])]
Out:
A B C D E
2017-04-01 0.522241 0.495106 -0.268194 -0.035003 one
2017-04-02 2.104572 -0.977768 -0.139632 -0.735926 one