Python 的 pandas 实践:
1 # !/usr/bin/env python
2 # encoding: utf-8
3 __author__ = 'Administrator'
4 import pandas as pd
5 import numpy as np
6 import matplotlib.pyplot as plt
7
8
9 #一、创建对象
10 #1. 通过传递一个list对象来创建一个Series,pandas会默认创建整型索引:
11 s=pd.Series([1,3,4,np.nan,6,8])
12 print(s)
13 # 0 1.0
14 # 1 3.0
15 # 2 4.0
16 # 3 NaN
17 # 4 6.0
18 # 5 8.0
19 # dtype: float64
20
21 #2.通过传递一个numpy array,时间索引以及列标签来创建一个DataFrame:
22 dates=pd.date_range('20180301',periods=6)
23 print(dates)
24 # DatetimeIndex(['2018-03-01', '2018-03-02', '2018-03-03', '2018-03-04',
25 # '2018-03-05', '2018-03-06'],
26 # dtype='datetime64[ns]', freq='D')
27 df=pd.DataFrame(np.random.randn(6,4),index=dates,columns=list('ABCD'))
28 # numpy.random.randn(d0, d1, …, dn)是从标准正态分布中返回一个或多个样本值。(可含负数)
29 # numpy.random.rand(d0, d1, …, dn)的随机样本位于[0, 1)中。
30 #P=numpy.random.rand(N,K) #随机生成一个 N行 K列的矩阵
31 print(df)
32 # A B C D
33 # 2018-03-01 -0.451506 -0.884044 -0.916664 -0.763684
34 # 2018-03-02 -0.463568 0.340688 -0.077484 -0.237660
35 # 2018-03-03 -1.533427 0.301283 0.268640 -0.011027
36 # 2018-03-04 1.036050 0.402203 0.485365 2.086525
37 # 2018-03-05 0.221578 -0.821756 -0.265241 0.277563
38 # 2018-03-06 1.774195 -0.288553 1.527936 0.119153
39
40 # '''
41
42 #3.通过传递一个能够被转换成类似序列结构的字典对象来创建一个DataFrame:
43 df2=pd.DataFrame({
44 'A':1.,
45 'B':pd.Timestamp('20180301'),
46 'C':pd.Series(1,index=list(range(4)),dtype='float32'),
47 'D':np.array([3]*4,dtype='int32'),
48 'E':pd.Categorical(["test","train","test","train"]),
49 'F':'foo'})
50 print(df2)
51 # A B C D E F
52 # 0 1.0 2018-03-01 1.0 3 test foo
53 # 1 1.0 2018-03-01 1.0 3 train foo
54 # 2 1.0 2018-03-01 1.0 3 test foo
55 # 3 1.0 2018-03-01 1.0 3 train foo
56
57 #4.查看不同列的数据类型:
58 print(df2.dtypes)
59 # A float64
60 # B datetime64[ns]
61 # C float32
62 # D int32
63 # E category
64 # F object
65 # dtype: object
66
67 #二、查看数据
68 #1. 查看dataframe中头部和尾部的行:
69 print(df.head())
70 # A B C D
71 # 2018-03-01 -0.250132 -1.403066 1.234990 -3.077763
72 # 2018-03-02 0.387496 -0.389183 0.186663 1.124608
73 # 2018-03-03 -0.105463 -0.230739 -0.227575 0.308565
74 # 2018-03-04 -1.703507 0.194876 1.790366 -0.561566
75 # 2018-03-05 -0.511609 0.695915 0.398392 0.107062
76 print(df.tail(3))
77 # A B C D
78 # 2018-03-04 0.704065 0.492649 0.533961 -1.518723
79 # 2018-03-05 2.192819 -0.508099 -0.173966 -0.401864
80 # 2018-03-06 -0.839634 -0.314676 -0.808266 -0.578229
81
82 #2.显示索引、列和底层的numpy数据:
83 print(df.index)
84 # DatetimeIndex(['2018-03-01', '2018-03-02', '2018-03-03', '2018-03-04',
85 # '2018-03-05', '2018-03-06'],
86 # dtype='datetime64[ns]', freq='D')
87 print(df.columns)
88 #Index(['A', 'B', 'C', 'D'], dtype='object')
89 print(df.values)
90 # [[ 1.65612186 -0.47932887 0.9673593 -0.63872414]
91 # [ 0.12229686 0.08831358 1.07344126 -0.12742276]
92 # [ 0.54654075 0.77281164 -0.6396787 0.1585142 ]
93 # [-0.70695944 -2.12273423 -0.24549759 -0.09530991]
94 # [ 2.66920788 0.6520858 1.72857641 -1.34418643]
95 # [ 1.87333346 -0.42716996 0.49558928 -1.47606701]]
96
97 #3. describe()函数对于数据的快速统计汇总:
98 print(df.describe())
99 # A B C D
100 # count 6.000000 6.000000 6.000000 6.000000
101 # mean 0.399068 0.339270 0.755588 -0.459344
102 # std 0.890360 1.011113 0.851783 1.759264
103 # min -1.002101 -0.806772 -0.333761 -2.411582
104 # 25% -0.087757 -0.400563 0.338822 -1.782221
105 # 50% 0.577418 0.244011 0.502612 -0.622453
106 # 75% 1.096592 0.941454 1.376095 0.433235
107 # max 1.281508 1.795854 1.910586 2.284103
108
109 #4. 对数据的转置:
110 print(df.T)
111 # 2018-03-01 2018-03-02 2018-03-03 2018-03-04 2018-03-05 2018-03-06
112 # A 0.843347 -0.906826 -0.528945 1.186650 -1.839152 -0.508169
113 # B -0.105481 2.084689 -1.106710 0.521137 0.741946 0.399700
114 # C -0.786144 0.269116 -0.180710 3.345385 1.310786 -0.204216
115 # D 0.453731 -0.243617 0.701440 2.541094 1.337923 -0.673128
116
117 #5. 按轴进行排序
118 print(df.sort_index(axis=1,ascending=False)) # axis = 0是按行进行操作, axis=1是按列进行操作; ascending=False是只递减,否则递增
119 # D C B A
120 # 2018-03-01 0.389294 -0.227394 0.649234 0.639820
121 # 2018-03-02 0.680265 0.466626 -1.940228 0.843753
122 # 2018-03-03 1.520800 0.570192 1.244427 -0.715080
123 # 2018-03-04 0.309068 -0.224222 -0.226254 1.416381
124 # 2018-03-05 -1.854131 -0.403245 -0.017054 0.840840
125 # 2018-03-06 -1.991173 1.275825 0.913996 1.561550
126
127 #6. 按值进行排序
128 # print(df.sort(column='B')) #?? AttributeError: 'DataFrame' object has no attribute 'sort'
129
130 #三、选择
131 # 虽然标准的Python/Numpy的选择和设置表达式都能够直接派上用场,
132 # 但是作为工程使用的代码,我们推荐使用经过优化的pandas数据访问方式: .at, .iat, .loc, .iloc 和 .ix
133 #(一)获取:
134 #1. 选择一个单独的列,这将会返回一个Series,等同于 df.A:
135 print(df['A'])
136 # 2018-03-01 0.156236
137 # 2018-03-02 -0.041257
138 # 2018-03-03 -0.970551
139 # 2018-03-04 -1.751839
140 # 2018-03-05 1.521352
141 # 2018-03-06 0.828690
142 # Freq: D, Name: A, dtype: float64
143
144 #2. 通过[]进行选择,这将会对行进行切片
145 print(df[0:3])
146 # A B C D
147 # 2018-03-01 -0.432011 0.697033 -3.028116 -0.217882
148 # 2018-03-02 -1.744071 0.647694 1.031179 -1.043985
149 # 2018-03-03 -0.673125 0.689913 0.648986 -1.471825
150 print(df['20180302':'20180304'])
151 # A B C D
152 # 2018-03-02 -0.803947 0.147807 -0.248534 0.496719
153 # 2018-03-03 -1.518123 0.376390 -0.793349 0.612074
154 # 2018-03-04 0.146634 0.506102 1.316693 -0.801691
155
156 #(二)通过标签选择:
157 #1. 使用标签来获取一个交叉的区域:
158 print(df.loc[dates[0]])
159 # A -1.593039
160 # B 0.400735
161 # C -0.870638
162 # D -0.551766
163 # Name: 2018-03-01 00:00:00, dtype: float64
164 #2. 通过标签来在多个轴上进行选择:
165 print(df.loc[:,['A','B']])
166 # A B
167 # 2018-03-01 0.326446 0.633246
168 # 2018-03-02 0.169674 0.892832
169 # 2018-03-03 -0.755691 -2.028912
170 # 2018-03-04 -1.005360 0.529193
171 # 2018-03-05 -0.457140 0.842211
172 # 2018-03-06 0.343157 0.879763
173
174 #3. 标签切片
175 print(df.loc['20180302':'20180304',['A','B']])
176 # A B
177 # 2018-03-02 0.197173 0.040377
178 # 2018-03-03 2.064367 1.112152
179 # 2018-03-04 0.888216 -0.591129
180
181 #4. 对于返回的对象进行维度缩减
182 print(df.loc['20180302',['A','B']])
183 # A -0.259955
184 # B -0.019266
185 # Name: 2018-03-02 00:00:00, dtype: float64
186
187 #5. 获取一个标量
188 print(df.loc[dates[0],'A']) #-0.313259346223
189
190 #6. 快速访问一个标量(与上一个方法等价)
191 print(df.at[dates[0],'A']) #-0.313259346223
192
193 #(三)通过位置选择:
194 #1. 通过传递数值进行位置选择(选择的是行)
195 print(df.iloc[3])
196 # A 1.661488
197 # B -1.175748
198 # C 0.642823
199 # D -0.491914
200 # Name: 2018-03-04 00:00:00, dtype: float64
201
202 #2. 通过数值进行切片,与numpy/python 中的情况类似
203 print(df.iloc[3:5,0:2]) #选择第3、第4行,第1、第2列
204 # A B
205 # 2018-03-04 0.492426 0.412712
206 # 2018-03-05 0.541252 -0.009380
207
208 #3. 通过制定一个位置的列表,与numpy/python中的情况类似
209 print(df.iloc[[1,2,4],[0,2]])
210 # A C
211 # 2018-03-02 -0.638074 1.794516
212 # 2018-03-03 -0.403471 -0.934373
213 # 2018-03-05 -1.309320 1.353276
214
215 #4. 对行进行切片
216 print(df.iloc[1:3,:])
217 # A B C D
218 # 2018-03-02 1.980513 -0.218688 2.627449 1.314947
219 # 2018-03-03 -0.532379 1.382092 -1.270961 0.722475
220
221 #5. 对列进行切片
222 print(df.iloc[:,1:3])
223 # B C
224 # 2018-03-01 0.332228 -1.682811
225 # 2018-03-02 -0.533398 -0.254960
226 # 2018-03-03 -0.926688 0.890513
227 # 2018-03-04 -0.448742 0.763850
228 # 2018-03-05 -0.841622 0.514873
229 # 2018-03-06 -1.346557 1.516414
230
231 #6. 获取特定的值
232 print(df.iloc[1,1]) #0.481882236461
233 print(df.iat[1,1]) #0.481882236461
234
235
236
237 #(四)布尔索引:
238 #1. 使用一个单独列的值来选择数据:
239 print(df[df.A>0])
240 # A B C D
241 # 2018-03-02 0.566243 1.510954 -0.898180 0.856439
242 # 2018-03-03 1.008447 -1.597226 -0.665134 -0.287472
243 # 2018-03-05 0.952498 -0.144979 0.620468 -0.830652
244
245 #2. 使用where操作来选择数据:
246 print(df[df>0])
247 # A B C D
248 # 2018-03-01 0.892660 NaN NaN NaN
249 # 2018-03-02 1.512600 NaN NaN 1.375527
250 # 2018-03-03 0.970026 1.184603 1.182990 NaN
251 # 2018-03-04 1.913993 NaN 0.914778 0.137170
252 # 2018-03-05 0.482589 NaN NaN 0.668817
253 # 2018-03-06 NaN 0.539344 0.142892 NaN
254
255 #3. 使用isin()方法来过滤:
256 df2=df.copy()
257 df2['E']=['one','one','two','three','four','three']
258 print(df2)
259 # A B C D E
260 # 2018-03-01 -1.138724 0.566583 0.338254 2.072839 one
261 # 2018-03-02 -0.366949 0.335546 1.653024 1.445071 one
262 # 2018-03-03 0.724615 1.715933 -0.754757 -1.452252 two
263 # 2018-03-04 -0.881962 -0.173858 -0.340868 -0.556665 three
264 # 2018-03-05 -2.126513 -0.113010 -0.796566 0.210673 four
265 # 2018-03-06 0.716490 0.223395 -1.428238 0.328406 three
266 print(df2[df2['E'].isin(['two','four'])])
267 # A B C D E
268 # 2018-03-03 -0.737833 -1.161520 0.897204 -0.029158 two
269 # 2018-03-05 1.072054 1.234587 0.935680 -1.284542 four
270
271
272
273 #(五)设置:
274 #1. 设置一个新的列:
275 s1=pd.Series([1,2,3,4,5,6],index=pd.date_range('20180302',periods=6))
276 print(s1)
277 # 2018-03-02 1
278 # 2018-03-03 2
279 # 2018-03-04 3
280 # 2018-03-05 4
281 # 2018-03-06 5
282 # 2018-03-07 6
283 # Freq: D, dtype: int64
284 df['F']=s1
285 print(df)
286 # A B C D F
287 # 2018-03-01 2.413592 -0.336264 0.165597 2.143270 NaN
288 # 2018-03-02 -1.921596 -2.100707 -0.454461 0.563247 1.0
289 # 2018-03-03 -0.235034 -0.517009 -2.409731 -0.711854 2.0
290 # 2018-03-04 0.667604 -0.838737 -0.425916 -0.238519 3.0
291 # 2018-03-05 1.057415 1.457143 0.440690 0.948613 4.0
292 # 2018-03-06 0.539187 -0.952633 0.316752 0.422146 5.0
293
294 #2. 通过标签设置新的值:
295 df.at[dates[0],'A']=0
296
297 #3. 通过位置设置新的值:
298 df.iat[0,1]=0
299
300 #4. 通过一个numpy数组设置一组新值:
301 df.loc[:,'D']=np.array([5]*len(df))
302 print(df)
303 # A B C D F
304 # 2018-03-01 0.000000 0.000000 0.164267 5 NaN
305 # 2018-03-02 0.614534 -0.865975 -0.977389 5 1.0
306 # 2018-03-03 -0.253095 -1.451951 2.360233 5 2.0
307 # 2018-03-04 0.143115 0.363544 1.587648 5 3.0
308 # 2018-03-05 0.010932 0.802590 -1.701589 5 4.0
309 # 2018-03-06 -0.354579 0.830066 0.404646 5 5.0
310
311 #5. 通过where操作来设置新的值:
312 df2=df.copy()
313 df2[df2>0]=-df2
314 print(df2)
315 # A B C D F
316 # 2018-03-01 0.000000 0.000000 -1.385454 -5 NaN
317 # 2018-03-02 -0.773506 -0.444692 -0.620307 -5 -1.0
318 # 2018-03-03 -0.506590 -2.445527 -0.664229 -5 -2.0
319 # 2018-03-04 -0.568711 -0.709224 -2.582502 -5 -3.0
320 # 2018-03-05 -1.074985 -2.480905 -0.537869 -5 -4.0
321 # 2018-03-06 -2.659346 -1.055430 -0.379758 -5 -5.0
322
323
324
325 #四、缺失值处理
326 # 在pandas中,使用np.nan来代替缺失值,这些值将默认不会包含在计算中,详情请参阅:Missing Data Section。
327 #1. reindex()方法可以对指定轴上的索引进行改变/增加/删除操作,这将返回原始数据的一个拷贝:
328 df1=df.reindex(index=dates[0:4],columns=list(df.columns)+['E'])
329 df1.loc[dates[0]:dates[1],'E']=1
330 print(df1)
331 # A B C D E
332 # 2018-03-01 -0.275255 -0.290044 0.707118 1.094318 1.0
333 # 2018-03-02 -1.340747 0.633546 -0.911210 -0.275105 1.0
334 # 2018-03-03 -1.044219 0.659945 1.370910 0.262282 NaN
335 # 2018-03-04 -0.015582 1.540852 -0.792882 -0.380751 NaN
336
337 #2. 去掉包含缺失值的行:
338 # df1=df1.dropna(how='any')
339 # print(df1)
340 # # A B C D E
341 # 2018-03-01 -0.914568 0.784980 -1.698139 -0.096874 1.0
342 # 2018-03-02 -0.410249 -0.494166 0.932946 -0.467547 1.0
343
344 #3. 对缺失值进行填充:
345 df1=df1.fillna(value=5)
346 print(df1)
347 # A B C D E
348 # 2018-03-01 -1.265605 0.778767 -0.947968 -1.330982 1.0
349 # 2018-03-02 1.778973 -1.428542 1.257860 0.362724 1.0
350 # 2018-03-03 -1.589094 -0.517478 -0.164942 -0.507224 5.0
351 # 2018-03-04 2.363145 2.089114 -0.081683 -0.184851 5.0
352
353 #4.对数据进行布尔填充
354 df1=pd.isnull(df1)
355 print(df1)
356 # A B C D E
357 # 2018-03-01 False False False False False
358 # 2018-03-02 False False False False False
359 # 2018-03-03 False False False False False
360 # 2018-03-04 False False False False False
361
362
363
364
365 #五、相关操作
366 # (一)统计(相关操作通常情况下不包括缺失值)
367 # #1. 执行描述性统计:
368 print(df.mean())
369 # A -0.066441
370 # B 0.154609
371 # C -0.154372
372 # D -0.155221
373 # dtype: float64
374
375 #2. 在其他轴上进行相同的操作:
376 print(df.mean(1))
377 # 2018-03-01 -0.138352
378 # 2018-03-02 -0.226558
379 # 2018-03-03 0.121705
380 # 2018-03-04 0.855662
381 # 2018-03-05 -0.892621
382 # 2018-03-06 0.062726
383 # Freq: D, dtype: float64
384
385 #3.对于拥有不同维度,需要对齐的对象进行操作。Pandas会自动的沿着指定的维度进行广播:
386
387
388 # (二)Apply
389 #1. 对数据应用函数:
390 print(df)
391 print(df.apply(np.cumsum))
392 # A B C D
393 # 2018-03-01 -0.381460 -0.296346 1.229803 -1.300226
394 # 2018-03-02 0.365891 0.974026 1.570268 -2.572981
395 # 2018-03-03 0.624070 0.211935 0.635084 -1.110378
396 # 2018-03-04 2.945062 -0.406832 -0.043918 -0.470773
397 # 2018-03-05 3.542080 0.092974 -1.585544 -0.658267
398 # 2018-03-06 3.440084 0.448828 -2.400617 -0.734055
399 print(df.apply(lambda x:x.max()-x.min()))
400 # A 2.702452
401 # B 2.032463
402 # C 2.771429
403 # D 2.762828
404 # dtype: float64
405
406 # (三)直方图
407 s=pd.Series(np.random.randint(0,7,size=10))
408 print(s)
409 # 0 2
410 # 1 6
411 # 2 6
412 # 3 3
413 # 4 3
414 # 5 4
415 # 6 4
416 # 7 6
417 # 8 6
418 # 9 2
419 # dtype: int32
420 print(s.value_counts())
421 # 6 4
422 # 4 2
423 # 3 2
424 # 2 2
425 # dtype: int64
426
427
428 # (四)字符串方法
429 # Series对象在其str属性中配备了一组字符串处理方法,可以很容易的应用到数组中的每个元素,如下段代码所示。
430 s=pd.Series(['A','B','C','Aaba','Baca',np.nan,'CABA','dog','cat'])
431 print(s.str.lower())
432 # 0 a
433 # 1 b
434 # 2 c
435 # 3 aaba
436 # 4 baca
437 # 5 NaN
438 # 6 caba
439 # 7 dog
440 # 8 cat
441 # dtype: object
442
443
444
445 #六、合并
446 #Pandas提供了大量的方法能够轻松的对Series,DataFrame和Panel对象进行各种符合各种逻辑关系的合并操作。
447 #1、Concat
448 df=pd.DataFrame(np.random.randn(10,4))
449 print(df)
450 # 0 1 2 3
451 # 0 0.620744 -0.921194 0.130483 -0.305914
452 # 1 0.311699 -0.085041 0.638297 -0.077868
453 # 2 0.327473 -0.732598 -0.134463 0.498805
454 # 3 -0.622715 -0.819375 -0.473504 -0.379117
455 # 4 -1.309207 -0.794917 -1.284665 0.830677
456 # 5 -1.170121 -2.063048 -0.836381 0.925829
457 # 6 -0.766342 0.454018 -0.181846 -1.052607
458 # 7 -0.996856 0.189226 0.428375 -1.149523
459 # 8 1.080517 1.884718 -0.065141 -0.781686
460 # 9 0.087353 0.209678 -1.333989 0.863220
461
462 #break it into pieces
463 pieces=[df[:3],df[3:7],df[7:]]
464 print(pieces)
465 print(pd.concat(pieces))
466 # 0 1 2 3
467 # 0 1.187009 -0.493550 0.777065 1.494107
468 # 1 -0.915190 1.228669 0.216910 1.610432
469 # 2 -0.647737 1.961472 1.369682 -1.195257
470 # 3 1.474973 1.968576 1.282678 -1.798167
471 # 4 1.449858 -1.828631 -0.217424 0.992141
472 # 5 -1.056223 0.464964 0.135468 0.181781
473 # 6 -1.677772 1.456419 0.642563 -0.895238
474 # 7 0.123780 0.030988 1.960217 0.140918
475 # 8 1.071418 1.737486 -0.170948 0.859271
476 # 9 -0.056640 -1.439686 -0.358960 -1.765060
477
478
479 #2、Join .类似于SQL类型的合并。
480 left=pd.DataFrame({'key':['foo','foo'],'lval':[1,2]})
481 print(left)
482 # key lval
483 # 0 foo 1
484 # 1 foo 2
485 right=pd.DataFrame({'key':['foo','foo'],'rval':[4,5]})
486 print(right)
487 # key rval
488 # 0 foo 4
489 # 1 foo 5
490 pd1=pd.merge(left,right,on='key')
491 print(pd1)
492 # key lval rval
493 # 0 foo 1 4
494 # 1 foo 1 5
495 # 2 foo 2 4
496 # 3 foo 2 5
497
498 #3、Append。将一行连接到一个DataFrame上。
499 df=pd.DataFrame(np.random.randn(8,4),columns=['A','B','C','D'])
500 print(df)
501 # A B C D
502 # 0 0.205671 -1.236797 -1.127111 1.422836
503 # 1 0.646151 0.202197 -0.160218 -0.839145
504 # 2 1.479783 -0.678455 0.649959 -1.085791
505 # 3 -0.851987 -0.821248 0.125836 0.819543
506 # 4 -1.312988 -0.898903 -0.420592 1.672173
507 # 5 0.240516 -0.711331 -0.717536 0.620066
508 # 6 -0.442280 0.539277 -1.428910 1.060193
509 # 7 0.257239 -2.034086 1.121833 1.518571
510 s=df.iloc[3]
511 df1=df.append(s,ignore_index=True)
512 print(df1)
513 # A B C D
514 # 0 0.205671 -1.236797 -1.127111 1.422836
515 # 1 0.646151 0.202197 -0.160218 -0.839145
516 # 2 1.479783 -0.678455 0.649959 -1.085791
517 # 3 -0.851987 -0.821248 0.125836 0.819543
518 # 4 -1.312988 -0.898903 -0.420592 1.672173
519 # 5 0.240516 -0.711331 -0.717536 0.620066
520 # 6 -0.442280 0.539277 -1.428910 1.060193
521 # 7 0.257239 -2.034086 1.121833 1.518571
522 # 8 -0.851987 -0.821248 0.125836 0.819543
523
524
525 #七、分组
526 #对于“group by”操作,我们通常是指以下一个或多个操作步骤:
527 # * (splitting)按照一些规则将数据分为不同的组;
528 # * (applying)对于每组数据分别执行一个函数;
529 # * (combining)将结果组合到一个数据结构中;
530
531 df=pd.DataFrame({'A':['foo','bar','foo','bar','foo','bar','foo','foo'],
532 'B':['one','one','two','three','two','two','one','three'],
533 'C':np.random.randn(8),
534 'D':np.random.randn(8) })
535 print(df)
536 # A B C D
537 # 0 foo one 0.792610 0.153922
538 # 1 bar one 1.497661 0.548711
539 # 2 foo two 0.038679 1.100214
540 # 3 bar three -1.074874 0.238335
541 # 4 foo two 1.176477 1.260415
542 # 5 bar two -0.629367 -1.098556
543 # 6 foo one 0.015918 -1.646855
544 # 7 foo three -0.486434 -0.930165
545
546 #1、分组并对每个分组执行sum函数:
547 dfg=df.groupby('A').sum()
548 print(dfg)
549 # C D
550 # A
551 # bar -0.20658 -0.311509
552 # foo 1.53725 -0.062469
553 #2、通过多个列进行分组形成一个层次索引,然后执行函数:
554 dfg2=df.groupby(['A','B']).sum()
555 print(dfg2)
556 # C D
557 # A B
558 # bar one 1.497661 0.548711
559 # three -1.074874 0.238335
560 # two -0.629367 -1.098556
561 # foo one 0.808528 -1.492933
562 # three -0.486434 -0.930165
563 # two 1.215156 2.360629
564
565 #八、Reshapeing
566 #1、Stack
567 tuples=list(zip(*[['bar','bar','baz','baz','foo','foo','quz','quz'],
568 ['one','two','one','two','one','two','one','two']]))
569 index=pd.MultiIndex.from_tuples(tuples,names=['first','second'])
570 df=pd.DataFrame(np.random.randn(8,2),index=index,columns=['A','B'])
571 df2=df[:4]
572 print(df2)
573 # A B
574 # first second
575 # bar one 1.146806 0.413660
576 # two -0.241280 -0.756498
577 # baz one -0.429149 -1.598932
578 # two 0.103805 -2.092773
579
580 stacked=df2.stack()
581 print(stacked)
582 # first second
583 # bar one A -0.671894
584 # B 0.488440
585 # two A -0.085894
586 # B -0.888060
587 # baz one A -0.647487
588 # B -1.573074
589 # two A 0.084324
590 # B -0.216785
591 # dtype: float64
592
593 stacked0=stacked.unstack()
594 print(stacked0)
595 # A B
596 # first second
597 # bar one -2.281352 0.683124
598 # two -2.555841 0.020481
599 # baz one 1.007699 -0.605463
600 # two 1.177308 0.833826
601 stacked1=stacked.unstack(1)
602 print(stacked1)
603 # second one two
604 # first
605 # bar A -2.281352 -2.555841
606 # B 0.683124 0.020481
607 # baz A 1.007699 1.177308
608 # B -0.605463 0.833826
609 stacked2=stacked.unstack(0)
610 print(stacked2)
611 # first bar baz
612 # second
613 # one A -0.279379 0.011654
614 # B 0.713347 0.482510
615 # two A -0.980093 0.536366
616 # B -0.378279 -1.023949
617
618 #2、数据透视表
619 df=pd.DataFrame({'A':['one','one','two','three']*3,
620 'B':['A','B','C']*4,
621 'C':['foo','foo','foo','bar','bar','bar']*2,
622 'D':np.random.randn(12),
623 'E':np.random.randn(12) })
624 print(df)
625 # A B C D E
626 # 0 one A foo -1.037929 -0.967839
627 # 1 one B foo 0.143201 1.936801
628 # 2 two C foo -1.108452 1.350176
629 # 3 three A bar 0.696497 0.578974
630 # 4 one B bar -1.206393 1.218049
631 # 5 one C bar -0.814728 0.440277
632 # 6 two A foo -2.039865 -1.298114
633 # 7 three B foo -0.155810 -0.249138
634 # 8 one C foo -0.436593 0.548266
635 # 9 one A bar -2.236853 -1.218478
636 # 10 two B bar -0.542738 -1.018322
637 # 11 three C bar -0.657995 -0.772053
638 #可以从这个数据中轻松的生成数据透视表:
639 pdtable=pd.pivot_table(df,values='D',index=['A','B'],columns=['C'])
640 print(pdtable)
641 # C bar foo
642 # A B
643 # one A 0.878124 0.739554
644 # B 1.508778 -0.261956
645 # C 0.452780 0.850025
646 # three A -0.616593 NaN
647 # B NaN -0.924248
648 # C -0.778909 NaN
649 # two A NaN -0.249317
650 # B 0.341066 NaN
651 # C NaN 0.706030
652 # '''
653 #九、时间序列
654 #Pandas在对频率转换进行重新采样时拥有简单、强大且高效的功能(如将按秒采样的数据转换为按5分钟为单位进行采样的数据)。这种操作在金融领域非常常见。
655 # rng=pd.date_range('1/1/2018',periods=100,freq='S')
656 # ts=pd.Series(np.random.randint(0,500,len(rng)),index=rng)
657 # ts0=ts.resample('5Min',how='sum')
658 # ........
659 # ........
660
661 #十、Categorical
662 #从0.15版本开始,pandas可以在DataFrame中支持Categorical类型的数据
663
664 #1、将原始的grade转换为Categorical数据类型:
665 # ........
666 # ........
667
668 #十一、画图
669 ts=pd.Series(np.random.randn(1000),index=pd.date_range('1/1/2018',periods=1000))
670 ts=ts.cumsum()
671 ts.plot()
672 # ........
673 # ........
674
675 #十二、导入和保存数据
676 #(一)CSV
677 #1、写入 csv文件
678 df.to_csv('foo.csv')
679 #2、从CSV文件中读取:
680 pd.read_csv('foo.csv')
681
682 #(二)HDF5
683 #1、
684 # ........
685 # ........
686
687 #(三)Excel
688 #1、写入excel文件:
689 df.to_excel('foo.xlsx',sheet_name='Sheet1')
690 #2、从excel文件中读取:
691 pd.read_excel('foo.xlsx','Sheet1',index_col=None,na_values=['NA'])
【Reference】
1、十分钟搞定pandas
2、10 Minutes to pandas