如何从索引匹配一天获取全天数据

Question

我有一个数据框 df1 按日期时间索引，数周内每分钟输入一次 示例：

           SAMPLE_TIME       Bottom     Top      Out     state                                                                    
0  2015-07-15 16:41:56      48.625   55.812   43.875        1       
1  2015-07-15 16:42:55      48.750   55.812   43.875        1     
2  2015-07-15 16:43:55      48.937   55.812   43.875        1       
3  2015-07-15 16:44:56      49.125   55.812   43.812        1      
4  2015-07-15 16:45:55      49.312   55.812   43.812        1     

我想找到 Avg(TempBottom,TempTop) 最低的那一天，然后按分钟获取全天数据，这样我就可以绘制那一天，我试过了：

df2 = df1.groupby(pd.TimeGrouper('D')).agg(min) \
.sort(['TempTop','TempBottom'], ascending=[True,True])

这给了我订购的最低温度天数。 示例：

SAMPLE_TIME       Bottom     Top      Out     state                                                                    
2015-10-17       19.994   25.840   21.875        0       
2015-08-29       26.182   28.777   25.937        0       
2015-11-19       19.244   33.027   28.937        0        
2015-11-07       19.744   33.527   28.125        0           

然后我只需要从 df2 获取第一个条目的索引：

 df1[df2.index[1]]

但我收到一个错误：

KeyError: Timestamp('2015-08-29 00:00:00')

Answer 1

来自docs：

警告

以下选择将引发KeyError；否则这种选择方法将与 pandas 中的其他选择方法不一致（因为这不是切片，也不会解析为切片）

dft['2013-1-15 12:30:00']

要选择单行，请使用.loc

In [71]: dft.loc['2013-1-15 12:30:00']
Out[71]: 
A    0.193284
Name: 2013-01-15 12:30:00, dtype: float64

所以你需要在你的情况下使用loc 方法：

In [103]: df1.loc[df2.index[0]]
Out[103]: 
           SAMPLE_TIME  TempBottom  TempTop  TempOut  State  Bypass
2015-07-15    16:41:56      48.625   55.812   43.875      1       1
2015-07-15    16:42:55      48.750   55.812   43.875      1       1
2015-07-15    16:43:55      48.937   55.812   43.875      1       1
2015-07-15    16:44:56      49.125   55.812   43.812      1       1
2015-07-15    16:45:55      49.312   55.812   43.812      1       1

编辑

当您传递单个参数时，它会尝试使用标签进行访问。但是，当您传递间隔时，它将用作切片。您可以通过技巧来传递价值 + 1 天：

In [276]: df2.index[0]
Out[276]: Timestamp('2015-07-15 00:00:00', offset='D')

In [277]: df2.index[0] + 1
Out[277]: Timestamp('2015-07-16 00:00:00', offset='D')

In [278]: df1.loc[df2.index[0]: df2.index[0] + 1]
Out[278]: 
                     TempBottom  TempTop  TempOut  State  Bypass
SAMPLE_TIME                                                     
2015-07-15 16:41:56      48.625   55.812   43.875      1       1
2015-07-15 16:42:55      48.750   55.812   43.875      1       1
2015-07-15 16:43:55      48.937   55.812   43.875      1       1
2015-07-15 16:44:56      49.125   55.812   43.812      1       1
2015-07-15 16:45:55      49.312   55.812   43.812      1       1

EDIT2

或者您可以将date 的Timestamp 转换为str：

In [355]: df2.index[0]
Out[355]: Timestamp('2015-07-15 00:00:00', offset='D')

In [356]: df2.index[0].date()
Out[356]: datetime.date(2015, 7, 15)

In [357]: str(df2.index[0].date())
Out[357]: '2015-07-15'

In [359]: df1[str(df2.index[0].date())]
Out[359]: 
                     TempBottom  TempTop  TempOut  State  Bypass
2015-07-15 16:41:56      48.625   55.812   43.875      1       1
2015-07-15 16:42:55      48.750   55.812   43.875      1       1
2015-07-15 16:43:55      48.937   55.812   43.875      1       1
2015-07-15 16:44:56      49.125   55.812   43.812      1       1
2015-07-15 16:45:55      49.312   55.812   43.812      1       1

Answer 2

所以这是我所做的思考过程，结合@Anton Protopopov答案：

In [1]: df1.ix[df2]
# call trace
ValueError: Cannot index with multidimensional key

In [2]: df1.ix[df2.index]
out[2]:
SAMPLE_TIME       Bottom     Top      Out     state                                                                    
2015-10-17          NaN      NaN      NaN      NaN        
2015-08-29          NaN      NaN      NaN      NaN         
2015-11-19          NaN      NaN      NaN      NaN        
2015-11-07          NaN      NaN      NaN      NaN         

In [3]: df1.ix[df2.index[4:5]]
Out[3]: 
SAMPLE_TIME       Bottom     Top      Out     state                                                                    
2015-11-04           NaN      NaN      NaN      NaN     

In [33]: df1.loc[df2.index[4:5]]
KeyError: "None of [DatetimeIndex(['2015-11-04'], dtype='datetime64[ns]', name=u'SAMPLE_TIME', freq=None, tz=None)] are in the [index]"

最后我放弃了ix 并决定让loc 工作，因为Anton 建议我尝试：

In [4]: df1.loc[df2.index[0].date()]
KeyError: 'the label [2015-11-04] is not in the [index]'

让我想到 loc 只接受最终有效的字符串：

In [5]: df1.loc[df2.index[4].strftime('%Y-%m-%d')]
Out[5]: 
SAMPLE_TIME              Bottom     Top      Out     state                                                                    
2015-11-04 00:00:22      56.256   56.300   43.750        0     
2015-11-04 00:01:22      56.256   56.300   43.812        0      
2015-11-04 00:02:22      56.256   56.300   43.812        0       
2015-11-04 00:03:22      56.256   56.300   43.812        0