背景
在数据分析中pandas
举足轻重,学习pandas
最好的方法就是看官方文档,以下是根据官方文档10 Minutes to pandas学习记录。(官方标题10分钟,感觉起码得半个小时吧)
在pandas
中主要有两种数据类型,可以简单的理解为:
- Series:一维数组
- DateFrame:二维数组(矩阵)
有了大概的概念之后,开始正式认识pandas
:
首先要引入对应的包:
import numpy as np
import pandas as pd
新建对象 Object Creation
-
Series
可以通过传入一个
list
对象来新建Series
,其中空值为np.nan
:s = pd.Series([1,3,4,np.nan,7,9]) s Out[5]: 0 1.0 1 3.0 2 4.0 3 NaN 4 7.0 5 9.0 dtype: float64
pandas
会默认创建一列索引index
(上面的0-5)。我们也可以在创建时就指定索引:pd.Series([1,3,4,np.nan,7,9], index=[1,1,2,2,'a',4]) Out[9]: 1 1.0 1 3.0 2 4.0 2 NaN a 7.0 4 9.0 dtype: float64
要注意的是,索引是可以重复的,也可以是字符。
-
DataFrame
新建一个DataFrame对象可以有多种方式:
-
通过传入一个numpy的数组、指定一个时间的索引以及一个列名。
dates = pd.date_range('20190101', periods=6) dates Out[11]: DatetimeIndex(['2019-01-01', '2019-01-02', '2019-01-03', '2019-01-04', '2019-01-05', '2019-01-06'], dtype='datetime64[ns]', freq='D') df = pd.DataFrame(np.random.randn(6,4), index=dates, columns=list('ABCD')) df Out[18]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 2019-01-06 0.221597 -0.753038 -1.741256 0.287280
-
通过传入一个dict对象
df2 = pd.DataFrame({'A':1., 'B':pd.Timestamp('20190101'), 'C':pd.Series(1, index=list(range(4)), dtype='float32'), 'D':np.array([3]*4, dtype='int32'), 'E':pd.Categorical(["test", "tain", "test", "train"]), 'F':'foo'}) df2 Out[27]: A B C D E F 0 1.0 2019-01-01 1.0 3 test foo 1 1.0 2019-01-01 1.0 3 tain foo 2 1.0 2019-01-01 1.0 3 test foo 3 1.0 2019-01-01 1.0 3 train foo
这里我们指定了不同的类型,可以通过如下查看:
df2.dtypes Out[28]: A float64 B datetime64[ns] C float32 D int32 E category F object dtype: object
-
可以看出DataFrame和Series一样,在没有指定索引时,会自动生成一个数字的索引,这在后续的操作中十分重要。
查看 Viewing Data
-
查看开头几行或者末尾几行:
df.head() Out[30]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 df.tail(3) Out[31]: A B C D 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 2019-01-06 0.221597 -0.753038 -1.741256 0.287280
可以通过添加行数参数来输出,默认为输出5行。
-
查看索引和列名
df.index Out[32]: DatetimeIndex(['2019-01-01', '2019-01-02', '2019-01-03', '2019-01-04', '2019-01-05', '2019-01-06'], dtype='datetime64[ns]', freq='D') df.columns Out[33]: Index(['A', 'B', 'C', 'D'], dtype='object')
-
使用
DataFrame.to_numpy()
转化为numpy
数据。需要注意的是由于numpy array
类型数据只可包含一种格式,而DataFrame
类型数据可包含多种格式,所以在转换过程中,pandas会找到一种可以处理DateFrame
中国所有格式的numpy array
格式,比如object
。这个过程会耗费一定的计算量。df.to_numpy() Out[35]: array([[ 0.67162219, 0.78572584, 0.39243527, 0.87469243], [-2.42070338, -1.11620768, -0.34607048, 0.78594081], [ 1.36442543, -0.94764138, 2.38688005, 0.58537186], [-0.48597971, -1.28145415, 0.35406263, -1.41885798], [-1.12271697, -2.78904135, -0.79181242, -0.17434484], [ 0.22159737, -0.75303807, -1.74125564, 0.28728004]]) df2.to_numpy() Out[36]: array([[1.0, Timestamp('2019-01-01 00:00:00'), 1.0, 3, 'test', 'foo'], [1.0, Timestamp('2019-01-01 00:00:00'), 1.0, 3, 'tain', 'foo'], [1.0, Timestamp('2019-01-01 00:00:00'), 1.0, 3, 'test', 'foo'], [1.0, Timestamp('2019-01-01 00:00:00'), 1.0, 3, 'train', 'foo']], dtype=object)
上面df全部为
float
类型,所以转换会很快,而df2涉及多种类型转换,最后全部变成了object
类型元素。 -
查看数据的简要统计结果
df.describe() Out[37]: A B C D count 6.000000 6.000000 6.000000 6.000000 mean -0.295293 -1.016943 0.042373 0.156680 std 1.356107 1.144047 1.396030 0.860725 min -2.420703 -2.789041 -1.741256 -1.418858 25% -0.963533 -1.240143 -0.680377 -0.058939 50% -0.132191 -1.031925 0.003996 0.436326 75% 0.559116 -0.801689 0.382842 0.735799 max 1.364425 0.785726 2.386880 0.874692
-
转置
df.T Out[38]: 2019-01-01 2019-01-02 2019-01-03 2019-01-04 2019-01-05 2019-01-06 A 0.671622 -2.420703 1.364425 -0.485980 -1.122717 0.221597 B 0.785726 -1.116208 -0.947641 -1.281454 -2.789041 -0.753038 C 0.392435 -0.346070 2.386880 0.354063 -0.791812 -1.741256 D 0.874692 0.785941 0.585372 -1.418858 -0.174345 0.287280
-
按坐标轴排序,其中axis参数为坐标轴,axis默认为0,即横轴(对行排序),axis=1则为纵轴(对列排序);asceding参数默认为True,即升序排序,ascending=False则为降序排序:
df.sort_index(axis=1) Out[44]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 df.sort_index(axis=1, ascending=False) Out[45]: D C B A 2019-01-01 0.874692 0.392435 0.785726 0.671622 2019-01-02 0.785941 -0.346070 -1.116208 -2.420703 2019-01-03 0.585372 2.386880 -0.947641 1.364425 2019-01-04 -1.418858 0.354063 -1.281454 -0.485980 2019-01-05 -0.174345 -0.791812 -2.789041 -1.122717 2019-01-06 0.287280 -1.741256 -0.753038 0.221597
可见
df.sort_index(axis=1)
是按列名升序排序,所以看起来没有变化,当设置ascending=False
时,列顺序变成了DCBA
。 -
按数值排序:
df.sort_values(by='B') Out[46]: A B C D 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 2019-01-01 0.671622 0.785726 0.392435 0.874692 df.sort_values(by='B', ascending=False) Out[47]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345
筛选 Selection
-
获取某列
df['A'] Out[49]: 2019-01-01 0.671622 2019-01-02 -2.420703 2019-01-03 1.364425 2019-01-04 -0.485980 2019-01-05 -1.122717 2019-01-06 0.221597 Freq: D, Name: A, dtype: float64 type(df.A) Out[52]: pandas.core.series.Series
也可直接用
df.A
,注意这里是大小写敏感的,这时候获取的是一个Series
类型数据。 -
选择多行
df[0:3] Out[53]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 df['20190102':'20190104'] Out[54]: A B C D 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858
通过一个
[]
会通过索引对行进行切片,由于前面设置了索引为日期格式,所以可以方便的直接使用日期范围进行筛选。 -
通过标签选择
-
选择某行
df.loc[dates[0]] Out[57]: A 0.671622 B 0.785726 C 0.392435 D 0.874692 Name: 2019-01-01 00:00:00, dtype: float64
-
选择指定行列的数据
df.loc[:, ('A', 'C')] Out[58]: A C 2019-01-01 0.671622 0.392435 2019-01-02 -2.420703 -0.346070 2019-01-03 1.364425 2.386880 2019-01-04 -0.485980 0.354063 2019-01-05 -1.122717 -0.791812 2019-01-06 0.221597 -1.741256 df.loc['20190102':'20190105', ('A', 'C')] Out[62]: A C 2019-01-02 -2.420703 -0.346070 2019-01-03 1.364425 2.386880 2019-01-04 -0.485980 0.354063 2019-01-05 -1.122717 -0.791812
传入第一个参数是行索引标签范围,第二个是列索引标签,
:
代表全部。 -
选定某值
df.loc['20190102', 'A'] Out[69]: -2.420703380445092 df.at[dates[1], 'A'] Out[70]: -2.420703380445092
可以通过
loc[]
和at[]
两种方式来获取某值,但需要注意的是,由于行索引为datetime
类型,使用loc[]
方式获取时,可直接使用20190102
字符串来代替,而在at[]
中,必须传入datetime
类型,否则会有报错:df.at['20190102', 'A'] File "pandas/_libs/index.pyx", line 81, in pandas._libs.index.IndexEngine.get_value File "pandas/_libs/index.pyx", line 89, in pandas._libs.index.IndexEngine.get_value File "pandas/_libs/index.pyx", line 449, in pandas._libs.index.DatetimeEngine.get_loc File "pandas/_libs/index.pyx", line 455, in pandas._libs.index.DatetimeEngine._date_check_type KeyError: '20190102'
-
-
通过位置选择
-
选择某行
df.iloc[3] Out[71]: A -0.485980 B -1.281454 C 0.354063 D -1.418858 Name: 2019-01-04 00:00:00, dtype: float64
iloc[]
方法的参数,必须是数值。 -
选择指定行列的数据
df.iloc[3:5, 0:2] Out[72]: A B 2019-01-04 -0.485980 -1.281454 2019-01-05 -1.122717 -2.789041 df.iloc[:,:] Out[73]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 df.iloc[[1, 2, 4], [0, 2]] Out[74]: A C 2019-01-02 -2.420703 -0.346070 2019-01-03 1.364425 2.386880 2019-01-05 -1.122717 -0.791812
同
loc[]
,:
代表全部。 -
选择某值
df.iloc[1, 1] Out[75]: -1.1162076820700824 df.iat[1, 1] Out[76]: -1.1162076820700824
可以通过
iloc[]
和iat[]
两种方法获取数值。
-
-
按条件判断选择
-
按某列的数值判断选择
df[df.A > 0] Out[77]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2019-01-06 0.221597 -0.753038 -1.741256 0.287280
-
筛选出符合要求的数据
df[df > 0] Out[78]: A B C D 2019-01-01 0.671622 0.785726 0.392435 0.874692 2019-01-02 NaN NaN NaN 0.785941 2019-01-03 1.364425 NaN 2.386880 0.585372 2019-01-04 NaN NaN 0.354063 NaN 2019-01-05 NaN NaN NaN NaN 2019-01-06 0.221597 NaN NaN 0.287280
不符合要求的数据均会被赋值为空
NaN
。 -
使用
isin()
方法筛选df2 = df.copy() df2['E'] = ['one', 'one', 'two', 'three', 'four', 'three'] df2 Out[88]: A B C D E 2019-01-01 0.671622 0.785726 0.392435 0.874692 one 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 one 2019-01-03 1.364425 -0.947641 2.386880 0.585372 two 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 three 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 four 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 three df2['E'].isin(['two', 'four']) Out[89]: 2019-01-01 False 2019-01-02 False 2019-01-03 True 2019-01-04 False 2019-01-05 True 2019-01-06 False Freq: D, Name: E, dtype: bool df2[df2['E'].isin(['two', 'four'])] Out[90]: A B C D E 2019-01-03 1.364425 -0.947641 2.386880 0.585372 two 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 four
注意:
isin
必须严格一致才行,df中的默认数值小数点位数很长,并非显示的5位,为了方便展示,所以新增了E列。直接用原数值,情况如下,可看出[1,1]
位置符合要求。df.isin([-1.1162076820700824]) Out[95]: A B C D 2019-01-01 False False False False 2019-01-02 False True False False 2019-01-03 False False False False 2019-01-04 False False False False 2019-01-05 False False False False 2019-01-06 False False False False
-
-
设定值
-
通过指定索引设定列
s1 = pd.Series([1, 2, 3, 4, 5, 6], index=pd.date_range('20190102', periods=6)) s1 Out[98]: 2019-01-02 1 2019-01-03 2 2019-01-04 3 2019-01-05 4 2019-01-06 5 2019-01-07 6 Freq: D, dtype: int64 df['F']=s1 df Out[101]: A B C D F 2019-01-01 0.671622 0.785726 0.392435 0.874692 NaN 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 1.0 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2.0 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 3.0 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 4.0 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 5.0
空值会自动填充为
NaN
。 -
通过标签设定值
df.at[dates[0], 'A'] = 0 df Out[103]: A B C D F 2019-01-01 0.000000 0.785726 0.392435 0.874692 NaN 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 1.0 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2.0 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 3.0 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 4.0 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 5.0
-
通过为止设定值
df.iat[0, 1] = 0 df Out[105]: A B C D F 2019-01-01 0.000000 0.000000 0.392435 0.874692 NaN 2019-01-02 -2.420703 -1.116208 -0.346070 0.785941 1.0 2019-01-03 1.364425 -0.947641 2.386880 0.585372 2.0 2019-01-04 -0.485980 -1.281454 0.354063 -1.418858 3.0 2019-01-05 -1.122717 -2.789041 -0.791812 -0.174345 4.0 2019-01-06 0.221597 -0.753038 -1.741256 0.287280 5.0
-
通过
NumPy array
设定值df.loc[:, 'D'] = np.array([5] * len(df)) df Out[109]: A B C D F 2019-01-01 0.000000 0.000000 0.392435 5 NaN 2019-01-02 -2.420703 -1.116208 -0.346070 5 1.0 2019-01-03 1.364425 -0.947641 2.386880 5 2.0 2019-01-04 -0.485980 -1.281454 0.354063 5 3.0 2019-01-05 -1.122717 -2.789041 -0.791812 5 4.0 2019-01-06 0.221597 -0.753038 -1.741256 5 5.0
-
通过条件判断设定值
df2 = df.copy() df2[df2 > 0] = -df2 df2 Out[112]: A B C D F 2019-01-01 0.000000 0.000000 -0.392435 -5 NaN 2019-01-02 -2.420703 -1.116208 -0.346070 -5 -1.0 2019-01-03 -1.364425 -0.947641 -2.386880 -5 -2.0 2019-01-04 -0.485980 -1.281454 -0.354063 -5 -3.0 2019-01-05 -1.122717 -2.789041 -0.791812 -5 -4.0 2019-01-06 -0.221597 -0.753038 -1.741256 -5 -5.0
-
空值处理 Missing Data
pandas
默认使用np.nan
来表示空值,在统计计算中会直接忽略。
通过reindex()
方法可以新增、修改、删除某坐标轴(行或列)的索引,并返回一个数据的拷贝:
df1 = df.reindex(index=dates[0:4], columns=list(df.columns) + ['E'])
df1.loc[dates[0]:dates[1], 'E'] = 1
df1
Out[115]:
A B C D F E
2019-01-01 0.000000 0.000000 0.392435 5 NaN 1.0
2019-01-02 -2.420703 -1.116208 -0.346070 5 1.0 1.0
2019-01-03 1.364425 -0.947641 2.386880 5 2.0 NaN
2019-01-04 -0.485980 -1.281454 0.354063 5 3.0 NaN
-
删除空值
df1.dropna(how='any') Out[116]: A B C D F E 2019-01-02 -2.420703 -1.116208 -0.34607 5 1.0 1.0
-
填充空值
df1.fillna(value=5) Out[117]: A B C D F E 2019-01-01 0.000000 0.000000 0.392435 5 5.0 1.0 2019-01-02 -2.420703 -1.116208 -0.346070 5 1.0 1.0 2019-01-03 1.364425 -0.947641 2.386880 5 2.0 5.0 2019-01-04 -0.485980 -1.281454 0.354063 5 3.0 5.0
-
判断是否为空值
pd.isna(df1) Out[118]: A B C D F E 2019-01-01 False False False False True False 2019-01-02 False False False False False False 2019-01-03 False False False False False True 2019-01-04 False False False False False True
运算 Operations
-
统计
注意 所有的统计默认是不包含空值的
-
平均值
默认情况是按列求平均值:
df.mean() Out[119]: A -0.407230 B -1.147897 C 0.042373 D 5.000000 F 3.000000 dtype: float64
如果需要按行求平均值,需指定轴参数:
df.mean(1) Out[120]: 2019-01-01 1.348109 2019-01-02 0.423404 2019-01-03 1.960733 2019-01-04 1.317326 2019-01-05 0.859286 2019-01-06 1.545461 Freq: D, dtype: float64
-
数值移动
s = pd.Series([1, 3, 5, np.nan, 6, 8], index=dates) s Out[122]: 2019-01-01 1.0 2019-01-02 3.0 2019-01-03 5.0 2019-01-04 NaN 2019-01-05 6.0 2019-01-06 8.0 Freq: D, dtype: float64 s = s.shift(2) s Out[125]: 2019-01-01 NaN 2019-01-02 NaN 2019-01-03 1.0 2019-01-04 3.0 2019-01-05 5.0 2019-01-06 NaN Freq: D, dtype: float64
这里将
s
的值移动两个,那么空出的部分会自动使用NaN
填充。 -
不同维度间的运算,
pandas
会自动扩展维度:df.sub(s, axis='index') Out[128]: A B C D F 2019-01-01 NaN NaN NaN NaN NaN 2019-01-02 NaN NaN NaN NaN NaN 2019-01-03 0.364425 -1.947641 1.386880 4.0 1.0 2019-01-04 -3.485980 -4.281454 -2.645937 2.0 0.0 2019-01-05 -6.122717 -7.789041 -5.791812 0.0 -1.0 2019-01-06 NaN NaN NaN NaN NaN
-
-
应用
通过
apply()
方法,可以对数据进行逐一操作:-
累计求和
df.apply(np.cumsum) Out[130]: A B C D F 2019-01-01 0.000000 0.000000 0.392435 5 NaN 2019-01-02 -2.420703 -1.116208 0.046365 10 1.0 2019-01-03 -1.056278 -2.063849 2.433245 15 3.0 2019-01-04 -1.542258 -3.345303 2.787307 20 6.0 2019-01-05 -2.664975 -6.134345 1.995495 25 10.0 2019-01-06 -2.443377 -6.887383 0.254239 30 15.0
这里使用了
apply()
方法调用np.cumsum
方法,也可直接使用df.cumsum()
:df.cumsum() Out[133]: A B C D F 2019-01-01 0.000000 0.000000 0.392435 5.0 NaN 2019-01-02 -2.420703 -1.116208 0.046365 10.0 1.0 2019-01-03 -1.056278 -2.063849 2.433245 15.0 3.0 2019-01-04 -1.542258 -3.345303 2.787307 20.0 6.0 2019-01-05 -2.664975 -6.134345 1.995495 25.0 10.0 2019-01-06 -2.443377 -6.887383 0.254239 30.0 15.0
-
自定义方法
通过自定义函数,配合
apply()
方法,可以实现更多数据处理:df.apply(lambda x: x.max() - x.min()) Out[134]: A 3.785129 B 2.789041 C 4.128136 D 0.000000 F 4.000000 dtype: float64
-
-
矩阵
统计矩阵中每个元素出现的频次:
s = pd.Series(np.random.randint(0, 7, size=10)) s Out[136]: 0 2 1 0 2 4 3 0 4 3 5 3 6 6 7 4 8 6 9 5 dtype: int64 s.value_counts() Out[137]: 6 2 4 2 3 2 0 2 5 1 2 1 dtype: int64
-
String方法
所有的Series类型都可以直接调用
str
的属性方法来对每个对象进行操作。-
比如转换成大写:
s = pd.Series(['A', 'B', 'C', 'Aaba', 'Baca', np.nan, 'CABA', 'dog', 'cat']) s.str.upper() Out[139]: 0 A 1 B 2 C 3 AABA 4 BACA 5 NaN 6 CABA 7 DOG 8 CAT dtype: object
-
分列:
s = pd.Series(['A,b', 'c,d']) s Out[142]: 0 A,b 1 c,d dtype: object s.str.split(',', expand=True) Out[143]: 0 1 0 A b 1 c d
-
其他方法:
dir(str) Out[140]: ['capitalize', 'casefold', 'center', 'count', 'encode', 'endswith', 'expandtabs', 'find', 'format', 'format_map', 'index', 'isalnum', 'isalpha', 'isascii', 'isdecimal', 'isdigit', 'isidentifier', 'islower', 'isnumeric', 'isprintable', 'isspace', 'istitle', 'isupper', 'join', 'ljust', 'lower', 'lstrip', 'maketrans', 'partition', 'replace', 'rfind', 'rindex', 'rjust', 'rpartition', 'rsplit', 'rstrip', 'split', 'splitlines', 'startswith', 'strip', 'swapcase', 'title', 'translate', 'upper', 'zfill']
-
合并 Merge
pandas`可以提供很多方法可以快速的合并各种类型的Series、DataFrame以及Panel Object。
-
Concat方法
df = pd.DataFrame(np.random.randn(10, 4)) df Out[145]: 0 1 2 3 0 -0.227408 -0.185674 -0.187919 0.185685 1 1.132517 -0.539992 1.156631 -0.022468 2 0.214134 -1.283055 -0.862972 0.518942 3 0.785903 1.033915 -0.471496 -1.403762 4 -0.676717 -0.529971 -1.161988 -1.265071 5 0.670126 1.320960 -0.128098 0.718631 6 0.589902 0.349386 0.221955 1.749188 7 -0.328885 0.607929 -0.973610 -0.928472 8 1.724243 -0.661503 -0.374254 0.409250 9 1.346625 0.618285 0.528776 -0.628470 # break it into pieces pieces = [df[:3], df[3:7], df[7:]] pieces Out[147]: [ 0 1 2 3 0 -0.227408 -0.185674 -0.187919 0.185685 1 1.132517 -0.539992 1.156631 -0.022468 2 0.214134 -1.283055 -0.862972 0.518942, 0 1 2 3 3 0.785903 1.033915 -0.471496 -1.403762 4 -0.676717 -0.529971 -1.161988 -1.265071 5 0.670126 1.320960 -0.128098 0.718631 6 0.589902 0.349386 0.221955 1.749188, 0 1 2 3 7 -0.328885 0.607929 -0.973610 -0.928472 8 1.724243 -0.661503 -0.374254 0.409250 9 1.346625 0.618285 0.528776 -0.628470] pd.concat(pieces) Out[148]: 0 1 2 3 0 -0.227408 -0.185674 -0.187919 0.185685 1 1.132517 -0.539992 1.156631 -0.022468 2 0.214134 -1.283055 -0.862972 0.518942 3 0.785903 1.033915 -0.471496 -1.403762 4 -0.676717 -0.529971 -1.161988 -1.265071 5 0.670126 1.320960 -0.128098 0.718631 6 0.589902 0.349386 0.221955 1.749188 7 -0.328885 0.607929 -0.973610 -0.928472 8 1.724243 -0.661503 -0.374254 0.409250 9 1.346625 0.618285 0.528776 -0.628470
-
Merge方法
这是类似
sql
的合并方法:left = pd.DataFrame({'key': ['foo', 'foo'], 'lval': [1, 2]}) right = pd.DataFrame({'key': ['foo', 'foo'], 'rval': [4, 5]}) left Out[151]: key lval 0 foo 1 1 foo 2 right Out[152]: key rval 0 foo 4 1 foo 5 pd.merge(left, right, on='key') Out[153]: key lval rval 0 foo 1 4 1 foo 1 5 2 foo 2 4 3 foo 2 5
另一个例子:
left = pd.DataFrame({'key': ['foo', 'bar'], 'lval': [1, 2]}) right = pd.DataFrame({'key': ['foo', 'bar'], 'rval': [4, 5]}) left Out[156]: key lval 0 foo 1 1 bar 2 right Out[157]: key rval 0 foo 4 1 bar 5 pd.merge(left, right, on='key') Out[158]: key lval rval 0 foo 1 4 1 bar 2 5
-
Append方法
在DataFrame中增加行
df = pd.DataFrame(np.random.randn(8, 4), columns=['A', 'B', 'C', 'D']) df Out[160]: A B C D 0 -0.496709 0.573449 0.076059 0.685285 1 0.479253 0.587376 -1.240070 -0.907910 2 -0.052609 -0.287786 -1.949402 1.163323 3 -0.659489 0.525583 0.820922 -1.368544 4 1.270453 -1.813249 0.059915 0.586703 5 1.859657 0.564274 -0.198763 -1.794173 6 -0.649153 -3.129258 0.063418 -0.727936 7 0.862402 -0.800031 -1.954784 -0.028607 s = df.iloc[3] s Out[162]: A -0.659489 B 0.525583 C 0.820922 D -1.368544 Name: 3, dtype: float64 df.append(s, ignore_index=True) Out[163]: A B C D 0 -0.496709 0.573449 0.076059 0.685285 1 0.479253 0.587376 -1.240070 -0.907910 2 -0.052609 -0.287786 -1.949402 1.163323 3 -0.659489 0.525583 0.820922 -1.368544 4 1.270453 -1.813249 0.059915 0.586703 5 1.859657 0.564274 -0.198763 -1.794173 6 -0.649153 -3.129258 0.063418 -0.727936 7 0.862402 -0.800031 -1.954784 -0.028607 8 -0.659489 0.525583 0.820922 -1.368544
这里要注意,我们增加了
ignore_index=True
参数,如果不设置的话,那么增加的新行的index
仍然是3
,这样在后续的处理中可能有存在问题。具体也需要看情况来处理。df.append(s) Out[164]: A B C D 0 -0.496709 0.573449 0.076059 0.685285 1 0.479253 0.587376 -1.240070 -0.907910 2 -0.052609 -0.287786 -1.949402 1.163323 3 -0.659489 0.525583 0.820922 -1.368544 4 1.270453 -1.813249 0.059915 0.586703 5 1.859657 0.564274 -0.198763 -1.794173 6 -0.649153 -3.129258 0.063418 -0.727936 7 0.862402 -0.800031 -1.954784 -0.028607 3 -0.659489 0.525583 0.820922 -1.368544
分组 Grouping
一般分组统计有三个步骤:
- 分组:选择需要的数据
- 计算:对每个分组进行计算
- 合并:把分组计算的结果合并为一个数据结构中
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)})
df
Out[166]:
A B C D
0 foo one -1.252153 0.172863
1 bar one 0.238547 -0.648980
2 foo two 0.756975 0.195766
3 bar three -0.933405 -0.320043
4 foo two -0.310650 -1.388255
5 bar two 1.568550 -1.911817
6 foo one -0.340290 -2.141259
按A列分组并使用sum
函数进行计算:
df.groupby('A').sum()
Out[167]:
C D
A
bar 0.873692 -2.880840
foo -1.817027 -5.833961
这里由于B列无法应用sum
函数,所以直接被忽略了。
按A、B列分组并使用sum
函数进行计算:
df.groupby(['A', 'B']).sum()
Out[168]:
C D
A B
bar one 0.238547 -0.648980
three -0.933405 -0.320043
two 1.568550 -1.911817
foo one -1.592443 -1.968396
three -0.670909 -2.673075
two 0.446325 -1.192490
这样就有了一个多层index的结果集。
整形 Reshaping
-
堆叠 Stack
python
的zip
函数可以将对象中对应的元素打包成一个个的元组:tuples = list(zip(['bar', 'bar', 'baz', 'baz', 'foo', 'foo', 'qux', 'qux'], ['one', 'two', 'one', 'two', 'one', 'two', 'one', 'two'])) tuples Out[172]: [('bar', 'one'), ('bar', 'two'), ('baz', 'one'), ('baz', 'two'), ('foo', 'one'), ('foo', 'two'), ('qux', 'one'), ('qux', 'two')] ## 设置两级索引 index = pd.MultiIndex.from_tuples(tuples, names=['first', 'second']) index Out[174]: MultiIndex(levels=[['bar', 'baz', 'foo', 'qux'], ['one', 'two']], codes=[[0, 0, 1, 1, 2, 2, 3, 3], [0, 1, 0, 1, 0, 1, 0, 1]], names=['first', 'second']) ## 创建DataFrame df = pd.DataFrame(np.random.randn(8, 2), index=index, columns=['A', 'B']) df Out[176]: A B first second bar one -0.501215 -0.947993 two -0.828914 0.232167 baz one 1.245419 1.006092 two 1.016656 -0.441073 foo one 0.479037 -0.500034 two -1.113097 0.591696 qux one -0.014760 -0.320735 two -0.648743 1.499899 ## 选取DataFrame df2 = df[:4] df2 Out[179]: A B first second bar one -0.501215 -0.947993 two -0.828914 0.232167 baz one 1.245419 1.006092 two 1.016656 -0.441073
使用
stack()
方法,可以通过堆叠的方式将二维数据变成为一维数据:stacked = df2.stack() stacked Out[181]: first second bar one A -0.501215 B -0.947993 two A -0.828914 B 0.232167 baz one A 1.245419 B 1.006092 two A 1.016656 B -0.441073 dtype: float64
对应的逆操作为
unstacked()
方法:stacked.unstack() Out[182]: A B first second bar one -0.501215 -0.947993 two -0.828914 0.232167 baz one 1.245419 1.006092 two 1.016656 -0.441073 stacked.unstack(1) Out[183]: second one two first bar A -0.501215 -0.828914 B -0.947993 0.232167 baz A 1.245419 1.016656 B 1.006092 -0.441073 stacked.unstack(0) Out[184]: first bar baz second one A -0.501215 1.245419 B -0.947993 1.006092 two A -0.828914 1.016656 B 0.232167 -0.441073
unstack()
默认对最后一层级进行操作,也可通过输入参数指定。 -
表格转置
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)}) df Out[190]: A B C D E 0 one A foo -0.933264 -2.387490 1 one B foo -0.288101 0.023214 2 two C foo 0.594490 0.418505 3 three A bar 0.450683 1.939623 4 one B bar 0.243897 -0.965783 5 one C bar -0.705494 -0.078283 6 two A foo 1.560352 0.419907 7 three B foo 0.199453 0.998711 8 one C foo 1.426861 -1.108297 9 one A bar -0.570951 -0.022560 10 two B bar -0.350937 -1.767804 11 three C bar 0.983465 0.065792
通过
pivot_table()
方法可以很方便的进行行列的转换:pd.pivot_table(df, values='D', index=['A', 'B'], columns=['C']) Out[191]: C bar foo A B one A -0.570951 -0.933264 B 0.243897 -0.288101 C -0.705494 1.426861 three A 0.450683 NaN B NaN 0.199453 C 0.983465 NaN two A NaN 1.560352 B -0.350937 NaN C NaN 0.594490
转换中,涉及到空值部分会自动填充为
NaN
。
时间序列 Time Series
pandas
的在时序转换方面十分强大,可以很方便的进行各种转换。
-
时间间隔调整
rng = pd.date_range('1/1/2019', periods=100, freq='S') rng[:5] Out[214]: DatetimeIndex(['2019-01-01 00:00:00', '2019-01-01 00:00:01', '2019-01-01 00:00:02', '2019-01-01 00:00:03', '2019-01-01 00:00:04'], dtype='datetime64[ns]', freq='S') ts = pd.Series(np.random.randint(0, 500, len(rng)), index=rng) ts.head(5) Out[216]: 2019-01-01 00:00:00 245 2019-01-01 00:00:01 347 2019-01-01 00:00:02 113 2019-01-01 00:00:03 196 2019-01-01 00:00:04 131 Freq: S, dtype: int64 ## 按10s间隔进行重新采样 ts1 = ts.resample('10S') ts1 Out[209]: DatetimeIndexResampler [freq=<10 * Seconds>, axis=0, closed=left, label=left, convention=start, base=0] ## 用求平均的方式进行数据整合 ts1.mean() Out[218]: 2019-01-01 00:00:00 174.0 2019-01-01 00:00:10 278.5 2019-01-01 00:00:20 281.8 2019-01-01 00:00:30 337.2 2019-01-01 00:00:40 221.0 2019-01-01 00:00:50 277.1 2019-01-01 00:01:00 171.0 2019-01-01 00:01:10 321.0 2019-01-01 00:01:20 318.6 2019-01-01 00:01:30 302.6 Freq: 10S, dtype: float64 ## 用求和的方式进行数据整合 ts1.sum() Out[219]: 2019-01-01 00:00:00 1740 2019-01-01 00:00:10 2785 2019-01-01 00:00:20 2818 2019-01-01 00:00:30 3372 2019-01-01 00:00:40 2210 2019-01-01 00:00:50 2771 2019-01-01 00:01:00 1710 2019-01-01 00:01:10 3210 2019-01-01 00:01:20 3186 2019-01-01 00:01:30 3026 Freq: 10S, dtype: int64
这里先通过
resample
进行重采样,在指定sum()
或者mean()
等方式来指定冲采样的处理方式。 -
显示时区:
rng = pd.date_range('1/1/2019 00:00', periods=5, freq='D') rng Out[221]: DatetimeIndex(['2019-01-01', '2019-01-02', '2019-01-03', '2019-01-04', '2019-01-05'], dtype='datetime64[ns]', freq='D') ts = pd.Series(np.random.randn(len(rng)), rng) ts Out[223]: 2019-01-01 -2.327686 2019-01-02 1.527872 2019-01-03 0.063982 2019-01-04 -0.213572 2019-01-05 -0.014856 Freq: D, dtype: float64 ts_utc = ts.tz_localize('UTC') ts_utc Out[225]: 2019-01-01 00:00:00+00:00 -2.327686 2019-01-02 00:00:00+00:00 1.527872 2019-01-03 00:00:00+00:00 0.063982 2019-01-04 00:00:00+00:00 -0.213572 2019-01-05 00:00:00+00:00 -0.014856 Freq: D, dtype: float64
-
转换时区:
ts_utc.tz_convert('US/Eastern') Out[226]: 2018-12-31 19:00:00-05:00 -2.327686 2019-01-01 19:00:00-05:00 1.527872 2019-01-02 19:00:00-05:00 0.063982 2019-01-03 19:00:00-05:00 -0.213572 2019-01-04 19:00:00-05:00 -0.014856 Freq: D, dtype: float64
-
时间格式转换
rng = pd.date_range('1/1/2019', periods=5, freq='M') ts = pd.Series(np.random.randn(len(rng)), index=rng) ts Out[230]: 2019-01-31 0.197134 2019-02-28 0.569082 2019-03-31 -0.322141 2019-04-30 0.005778 2019-05-31 -0.082306 Freq: M, dtype: float64 ps = ts.to_period() ps Out[232]: 2019-01 0.197134 2019-02 0.569082 2019-03 -0.322141 2019-04 0.005778 2019-05 -0.082306 Freq: M, dtype: float64 ps.to_timestamp() Out[233]: 2019-01-01 0.197134 2019-02-01 0.569082 2019-03-01 -0.322141 2019-04-01 0.005778 2019-05-01 -0.082306 Freq: MS, dtype: float64
在是时间段和时间转换过程中,有一些很方便的算术方法可以使用,比如我们转换如下两个频率:
1、按季度划分,且每个年的最后一个月是11月。
2、按季度划分,每个月开始为频率一中下一个月的早上9点。
prng = pd.period_range('2018Q1', '2019Q4', freq='Q-NOV') prng Out[243]: PeriodIndex(['2018Q1', '2018Q2', '2018Q3', '2018Q4', '2019Q1', '2019Q2', '2019Q3', '2019Q4'], dtype='period[Q-NOV]', freq='Q-NOV') ts = pd.Series(np.random.randn(len(prng)), prng) ts Out[245]: 2018Q1 -0.112692 2018Q2 -0.507304 2018Q3 -0.324846 2018Q4 0.549671 2019Q1 -0.897732 2019Q2 1.130070 2019Q3 -0.399814 2019Q4 0.830488 Freq: Q-NOV, dtype: float64 ts.index = (prng.asfreq('M', 'e') + 1).asfreq('H', 's') + 9 ts Out[247]: 2018-03-01 09:00 -0.112692 2018-06-01 09:00 -0.507304 2018-09-01 09:00 -0.324846 2018-12-01 09:00 0.549671 2019-03-01 09:00 -0.897732 2019-06-01 09:00 1.130070 2019-09-01 09:00 -0.399814 2019-12-01 09:00 0.830488 Freq: H, dtype: float64
注意:这个例子有点怪。可以这样理解,我们先将
prng
直接转换为按小时显示:prng.asfreq('H', 'end') Out[253]: PeriodIndex(['2018-02-28 23:00', '2018-05-31 23:00', '2018-08-31 23:00', '2018-11-30 23:00', '2019-02-28 23:00', '2019-05-31 23:00', '2019-08-31 23:00', '2019-11-30 23:00'], dtype='period[H]', freq='H')
我们要把时间转换为下一个月的早上9点,所以先转换为按月显示,并每个月加1(即下个月),然后按小时显示并加9(早上9点)。
另外例子中
s
参数是start
的简写,e
参数是end
的简写,Q-NOV
即表示按季度,且每年的NOV是最后一个月。更多了
freq
简称可以参考:http://pandas.pydata.org/pand...asfreq()
方法介绍可参考:http://pandas.pydata.org/pand...
分类目录类型 Categoricals
关于Categories类型介绍可以参考:http://pandas.pydata.org/pand...
-
类型转换:
astype('category')
df = pd.DataFrame({"id": [1, 2, 3, 4, 5, 6], "raw_grade": ['a', 'b', 'b', 'a', 'a', 'e']}) df Out[255]: id raw_grade 0 1 a 1 2 b 2 3 b 3 4 a 4 5 a 5 6 e df['grade'] = df['raw_grade'].astype('category') df['grade'] Out[257]: 0 a 1 b 2 b 3 a 4 a 5 e Name: grade, dtype: category Categories (3, object): [a, b, e]
-
重命名分类:
cat
df["grade"].cat.categories = ["very good", "good", "very bad"] df['grade'] Out[269]: 0 very good 1 good 2 good 3 very good 4 very good 5 very bad Name: grade, dtype: category Categories (3, object): [very good, good, very bad]
-
重分类:
df['grade'] = df['grade'].cat.set_categories(["very bad", "bad", "medium","good", "very good"]) df['grade'] Out[271]: 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
df.sort_values(by="grade") Out[272]: 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
-
分组
df.groupby("grade").size() Out[273]: grade very bad 1 bad 0 medium 0 good 2 very good 3 dtype: int64
画图 Plotting
-
Series
ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2000', periods=1000)) ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2019', periods=1000)) ts = ts.cumsum() ts.plot() Out[277]:
import matplotlib.pyplot as plt plt.show() -
DataFrame画图
使用
plot
可以把所有的列都通过标签的形式展示出来:df = pd.DataFrame(np.random.randn(1000, 4), index=ts.index, columns=['A', 'B', 'C', 'D']) df = df.cumsum() plt.figure() Out[282]:
导入导出数据 Getting Data In/Out
-
CSV
-
写入:
df.to_csv('foo.csv')
-
读取:
pd.read_csv('foo.csv')
-
-
HDF5
-
写入:
df.to_hdf('foo.h5', 'df')
-
读取:
pd.read_hdf('foo.h5', 'df')
-
-
Excel
-
写入:
df.to_excel('foo.xlsx', sheet_name='Sheet1')
-
读取:
pd.read_excel('foo.xlsx', 'Sheet1', index_col=None, na_values=['NA'])
-
异常处理 Gotchas
如果有一些异常情况比如:
>>> 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().
可以参考如下链接:
http://pandas.pydata.org/pand...
http://pandas.pydata.org/pand...