简介
pandas是建立在Python编程语言之上的一种快速,强大,灵活且易于使用的开源数据分析和处理工具,它含有使数据清洗和分析⼯作变得更快更简单的数据结构和操作⼯具。pandas经常和其它⼯具⼀同使⽤,如数值计算⼯具NumPy和SciPy,分析库statsmodels和scikit-learn,和数据可视化库matplotlib等。
pandas是基于NumPy数组构建的,虽然pandas采⽤了⼤量的NumPy编码⻛格,但⼆者最⼤的不同是pandas是专⻔为处理表格和混杂数据设计的。⽽NumPy更适合处理统⼀的数值数组数据。
本文是关于Pandas的简洁教程。
对象创建因为Pandas是基于NumPy数组来构建的,所以我们在引用的时候需要同时引用Pandas和NumPy:
In [1]: import numpy as np
In [2]: import pandas as pd
Pandas中最主要的两个数据结构是Series和DataFrame。
Series和一维数组很相似,它是由NumPy的各种数据类型来组成的,同时还包含了和这组数据相关的index。
我们来看一个Series的例子:
In [3]: pd.Series([1, 3, 5, 6, 8])
Out[3]:
0 1
1 3
2 5
3 6
4 8
dtype: int64
左边的是索引,右边的是值。因为我们在创建Series的时候并没有指定index,所以index是从0开始到n-1结束。
Series在创建的时候还可以传入np.nan表示空值:
In [4]: pd.Series([1, 3, 5, np.nan, 6, 8])
Out[4]:
0 1.0
1 3.0
2 5.0
3 NaN
4 6.0
5 8.0
dtype: float64
DataFrame是⼀个表格型的数据结构,它含有⼀组有序的列,每列可以是不同的值类型(数值、字符串、布尔值等)。
DataFrame既有⾏索引也有列索引,它可以被看做由Series组成的字典(共⽤同⼀个索引)。
看一个创建DataFrame的例子:
In [5]: dates = pd.date_range('20201201', periods=6)
In [6]: dates
Out[6]:
DatetimeIndex(['2020-12-01', '2020-12-02', '2020-12-03', '2020-12-04',
'2020-12-05', '2020-12-06'],
dtype='datetime64[ns]', freq='D')
上面我们创建了一个index的list。
然后使用这个index来创建一个DataFrame:
In [7]: pd.DataFrame(np.random.randn(6, 4), index=dates, columns=list('ABCD'))
Out[7]:
A B C D
2020-12-01 1.536312 -0.318095 -0.737956 0.143352
2020-12-02 1.325221 0.065641 -2.763370 -0.130511
2020-12-03 -1.143560 -0.805807 0.174722 0.427027
2020-12-04 -0.724206 0.050155 -0.648675 -0.645166
2020-12-05 0.182411 0.956385 0.349465 -0.484040
2020-12-06 1.857108 1.245928 -0.767316 -1.890586
上面的DataFrame接收三个参数,第一个参数是DataFrame的表格数据,第二个参数是index的值,也可以看做是行名,第三个参数是列名。
还可以直接传入一个字典来创建一个DataFrame:
In [9]: pd.DataFrame({'A': 1.,
...: 'B': pd.Timestamp('20201202'),
...: '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'})
...:
Out[9]:
A B C D E F
0 1.0 2020-12-02 1.0 3 test foo
1 1.0 2020-12-02 1.0 3 train foo
2 1.0 2020-12-02 1.0 3 test foo
3 1.0 2020-12-02 1.0 3 train foo
上面的DataFrame中,每个列都有不同的数据类型。
我们用个图片来更好的理解DataFrame和Series:
它就像是Excel中的表格,带有行头和列头。
DataFrame中的每一列都可以看做是一个Series:
查看数据创建好Series和DataFrame之后,我们就可以查看他们的数据了。
Series可以通过index和values来获取其索引和值信息:
In [10]: data1 = pd.Series([1, 3, 5, np.nan, 6, 8])
In [12]: data1.index
Out[12]: RangeIndex(start=0, stop=6, step=1)
In [14]: data1.values
Out[14]: array([ 1., 3., 5., nan, 6., 8.])
DataFrame可以看做是Series的集合,所以DataFrame带有更多的属性:
In [16]: df.head()
Out[16]:
A B C D
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599
2020-12-03 0.610846 0.216937 0.821258 0.805818
2020-12-04 0.490105 0.732421 0.547129 -0.443274
2020-12-05 -0.475531 -0.853141 0.160017 0.986973
In [17]: df.tail(3)
Out[17]:
A B C D
2020-12-04 0.490105 0.732421 0.547129 -0.443274
2020-12-05 -0.475531 -0.853141 0.160017 0.986973
2020-12-06 0.288091 -2.164323 0.193989 -0.197923
head跟tail分别取得DataFrame的头几行和尾部几行。
同样的DataFrame也有index和columns:
In [19]: df.index
Out[19]:
DatetimeIndex(['2020-12-01', '2020-12-02', '2020-12-03', '2020-12-04',
'2020-12-05', '2020-12-06'],
dtype='datetime64[ns]', freq='D')
In [20]: df.values
Out[20]:
array([[ 0.44624818, -0.0605494 , -0.44566462, -1.39250227],
[-1.11974917, -1.65977552, -0.61865617, 1.97159943],
[ 0.61084596, 0.2169369 , 0.82125808, 0.80581847],
[ 0.49010504, 0.73242082, 0.54712889, -0.44327351],
[-0.47553134, -0.85314134, 0.16001748, 0.98697257],
[ 0.28809148, -2.16432292, 0.19398863, -0.19792266]])
describe方法可以对数据进行统计:
In [26]: df.describe()
Out[26]:
A B C D
count 6.000000 6.000000 6.000000 6.000000
mean 0.040002 -0.631405 0.109679 0.288449
std 0.687872 1.128019 0.556099 1.198847
min -1.119749 -2.164323 -0.618656 -1.392502
25% -0.284626 -1.458117 -0.294244 -0.381936
50% 0.367170 -0.456845 0.177003 0.303948
75% 0.479141 0.147565 0.458844 0.941684
max 0.610846 0.732421 0.821258 1.971599
还可以对DataFrame进行转置:
In [27]: df.T
Out[27]:
2020-12-01 2020-12-02 2020-12-03 2020-12-04 2020-12-05 2020-12-06
A 0.446248 -1.119749 0.610846 0.490105 -0.475531 0.288091
B -0.060549 -1.659776 0.216937 0.732421 -0.853141 -2.164323
C -0.445665 -0.618656 0.821258 0.547129 0.160017 0.193989
D -1.392502 1.971599 0.805818 -0.443274 0.986973 -0.197923
可以按行和按列进行排序:
In [28]: df.sort_index(axis=1, ascending=False)
Out[28]:
D C B A
2020-12-01 -1.392502 -0.445665 -0.060549 0.446248
2020-12-02 1.971599 -0.618656 -1.659776 -1.119749
2020-12-03 0.805818 0.821258 0.216937 0.610846
2020-12-04 -0.443274 0.547129 0.732421 0.490105
2020-12-05 0.986973 0.160017 -0.853141 -0.475531
2020-12-06 -0.197923 0.193989 -2.164323 0.288091
In [29]: df.sort_values(by='B')
Out[29]:
A B C D
2020-12-06 0.288091 -2.164323 0.193989 -0.197923
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599
2020-12-05 -0.475531 -0.853141 0.160017 0.986973
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502
2020-12-03 0.610846 0.216937 0.821258 0.805818
2020-12-04 0.490105 0.732421 0.547129 -0.443274
选择数据
通过DataFrame的列名,可以选择代表列的Series:
In [30]: df['A']
Out[30]:
2020-12-01 0.446248
2020-12-02 -1.119749
2020-12-03 0.610846
2020-12-04 0.490105
2020-12-05 -0.475531
2020-12-06 0.288091
Freq: D, Name: A, dtype: float64
通过切片可以选择行:
In [31]: df[0:3]
Out[31]:
A B C D
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599
2020-12-03 0.610846 0.216937 0.821258 0.805818
或者这样:
In [32]: df['20201202':'20201204']
Out[32]:
A B C D
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599
2020-12-03 0.610846 0.216937 0.821258 0.805818
2020-12-04 0.490105 0.732421 0.547129 -0.443274
loc和iloc
使用loc可以使用轴标签来选取数据。
In [33]: df.loc[:, ['A', 'B']]
Out[33]:
A B
2020-12-01 0.446248 -0.060549
2020-12-02 -1.119749 -1.659776
2020-12-03 0.610846 0.216937
2020-12-04 0.490105 0.732421
2020-12-05 -0.475531 -0.853141
2020-12-06 0.288091 -2.164323
前面是行的选择,后面是列的选择。
还可以指定index的名字:
In [34]: df.loc['20201202':'20201204', ['A', 'B']]
Out[34]:
A B
2020-12-02 -1.119749 -1.659776
2020-12-03 0.610846 0.216937
2020-12-04 0.490105 0.732421
如果index的名字不是切片的话,将会给数据降维:
In [35]: df.loc['20201202', ['A', 'B']]
Out[35]:
A -1.119749
B -1.659776
Name: 2020-12-02 00:00:00, dtype: float64
如果后面列是一个常量的话,直接返回对应的值:
In [37]: df.loc['20201202', 'A']
Out[37]: -1.1197491665145112
iloc是根据值来选取数据,比如我们选择第三行:
In [42]: df.iloc[3]
Out[42]:
A 0.490105
B 0.732421
C 0.547129
D -0.443274
Name: 2020-12-04 00:00:00, dtype: float64
它其实和df.loc[‘2020-12-04’]是等价的:
In [41]: df.loc['2020-12-04']
Out[41]:
A 0.490105
B 0.732421
C 0.547129
D -0.443274
Name: 2020-12-04 00:00:00, dtype: float64
同样可以传入切片:
In [43]: df.iloc[3:5, 0:2]
Out[43]:
A B
2020-12-04 0.490105 0.732421
2020-12-05 -0.475531 -0.853141
可以传入list:
In [44]: df.iloc[[1, 2, 4], [0, 2]]
Out[44]:
A C
2020-12-02 -1.119749 -0.618656
2020-12-03 0.610846 0.821258
2020-12-05 -0.475531 0.160017
取具体某个格子的值:
In [45]: df.iloc[1, 1]
Out[45]: -1.6597755161871708
布尔索引
DataFrame还可以通过布尔值来进行索引,下面是找出列A中所有元素大于0的:
In [46]: df[df['A'] > 0]
Out[46]:
A B C D
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502
2020-12-03 0.610846 0.216937 0.821258 0.805818
2020-12-04 0.490105 0.732421 0.547129 -0.443274
2020-12-06 0.288091 -2.164323 0.193989 -0.197923
或者找出整个DF中,值大于0的:
In [47]: df[df > 0]
Out[47]:
A B C D
2020-12-01 0.446248 NaN NaN NaN
2020-12-02 NaN NaN NaN 1.971599
2020-12-03 0.610846 0.216937 0.821258 0.805818
2020-12-04 0.490105 0.732421 0.547129 NaN
2020-12-05 NaN NaN 0.160017 0.986973
2020-12-06 0.288091 NaN 0.193989 NaN
可以给DF添加一列:
In [48]: df['E'] = ['one', 'one', 'two', 'three', 'four', 'three']
In [49]: df
Out[49]:
A B C D E
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502 one
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599 one
2020-12-03 0.610846 0.216937 0.821258 0.805818 two
2020-12-04 0.490105 0.732421 0.547129 -0.443274 three
2020-12-05 -0.475531 -0.853141 0.160017 0.986973 four
2020-12-06 0.288091 -2.164323 0.193989 -0.197923 three
使用isin()来进行范围值的判断判断:
In [50]: df[df['E'].isin(['two', 'four'])]
Out[50]:
A B C D E
2020-12-03 0.610846 0.216937 0.821258 0.805818 two
2020-12-05 -0.475531 -0.853141 0.160017 0.986973 four
处理缺失数据
现在我们的df有a,b,c,d,e这5列,如果我们再给他加一列f,那么f的初始值将会是NaN:
In [55]: df.reindex(columns=list(df.columns) + ['F'])
Out[55]:
A B C D E F
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502 one NaN
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599 one NaN
2020-12-03 0.610846 0.216937 0.821258 0.805818 two NaN
2020-12-04 0.490105 0.732421 0.547129 -0.443274 three NaN
2020-12-05 -0.475531 -0.853141 0.160017 0.986973 four NaN
2020-12-06 0.288091 -2.164323 0.193989 -0.197923 three NaN
我们给前面的两个F赋值:
In [74]: df1.iloc[0:2,5]=1
In [75]: df1
Out[75]:
A B C D E F
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502 one 1.0
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599 one 1.0
2020-12-03 0.610846 0.216937 0.821258 0.805818 two NaN
2020-12-04 0.490105 0.732421 0.547129 -0.443274 three NaN
2020-12-05 -0.475531 -0.853141 0.160017 0.986973 four NaN
2020-12-06 0.288091 -2.164323 0.193989 -0.197923 three NaN
可以drop所有为NaN的行:
In [76]: df1.dropna(how='any')
Out[76]:
A B C D E F
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502 one 1.0
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599 one 1.0
可以填充NaN的值:
In [77]: df1.fillna(value=5)
Out[77]:
A B C D E F
2020-12-01 0.446248 -0.060549 -0.445665 -1.392502 one 1.0
2020-12-02 -1.119749 -1.659776 -0.618656 1.971599 one 1.0
2020-12-03 0.610846 0.216937 0.821258 0.805818 two 5.0
2020-12-04 0.490105 0.732421 0.547129 -0.443274 three 5.0
2020-12-05 -0.475531 -0.853141 0.160017 0.986973 four 5.0
2020-12-06 0.288091 -2.164323 0.193989 -0.197923 three 5.0
可以对值进行判断:
In [78]: pd.isna(df1)
Out[78]:
A B C D E F
2020-12-01 False False False False False False
2020-12-02 False False False False False False
2020-12-03 False False False False False True
2020-12-04 False False False False False True
2020-12-05 False False False False False True
2020-12-06 False False False False False True
合并
DF可以使用Concat来合并多个df,我们先创建一个df:
In [79]: df = pd.DataFrame(np.random.randn(10, 4))
In [80]: df
Out[80]:
0 1 2 3
0 1.089041 2.010142 -0.532527 0.991669
1 1.303678 -0.614206 -1.358952 0.006290
2 -2.663938 0.600209 -0.008845 -0.036900
3 0.863718 -0.450501 1.325427 0.417345
4 0.789239 -0.492630 0.873732 0.375941
5 0.327177 0.010719 -0.085967 -0.591267
6 -0.014350 1.372144 -0.688845 0.422701
7 -3.355685 0.044306 -0.979253 -2.184240
8 -0.051961 0.649734 1.156918 -0.233725
9 -0.692530 0.057805 -0.030565 0.209416
然后把DF拆成三部分:
In [81]: pieces = [df[:3], df[3:7], df[7:]]
最后把使用concat把他们合起来:
In [82]: pd.concat(pieces)
Out[82]:
0 1 2 3
0 1.089041 2.010142 -0.532527 0.991669
1 1.303678 -0.614206 -1.358952 0.006290
2 -2.663938 0.600209 -0.008845 -0.036900
3 0.863718 -0.450501 1.325427 0.417345
4 0.789239 -0.492630 0.873732 0.375941
5 0.327177 0.010719 -0.085967 -0.591267
6 -0.014350 1.372144 -0.688845 0.422701
7 -3.355685 0.044306 -0.979253 -2.184240
8 -0.051961 0.649734 1.156918 -0.233725
9 -0.692530 0.057805 -0.030565 0.209416
还可以使用join来进行类似SQL的合并:
In [83]: left = pd.DataFrame({'key': ['foo', 'foo'], 'lval': [1, 2]})
In [84]: right = pd.DataFrame({'key': ['foo', 'foo'], 'rval': [4, 5]})
In [85]: left
Out[85]:
key lval
0 foo 1
1 foo 2
In [86]: right
Out[86]:
key rval
0 foo 4
1 foo 5
In [87]: pd.merge(left, right, on='key')
Out[87]:
key lval rval
0 foo 1 4
1 foo 1 5
2 foo 2 4
3 foo 2 5
分组
先看上面的DF:
In [99]: df2
Out[99]:
key lval rval
0 foo 1 4
1 foo 1 5
2 foo 2 4
3 foo 2 5
我们可以根据key来进行group,从而进行sum:
In [98]: df2.groupby('key').sum()
Out[98]:
lval rval
key
foo 6 18
group还可以按多个列进行:
In [100]: df2.groupby(['key','lval']).sum()
Out[100]:
rval
key lval
foo 1 9
2 9
最通俗的解读,最深刻的干货,最简洁的教程,众多你不知道的小技巧等你来发现!