python - Pandas 重采样不能正常工作

Question

我从 pandas 那里得到了一个奇怪的行为，我想将我的分钟数据重新采样为每小时数据(使用平均值)。我的数据如下所示:

Data.head()
                      AAA    BBB
Time                              
2009-02-10 09:31:00  86.34  101.00
2009-02-10 09:36:00  86.57  100.50
2009-02-10 09:38:00  86.58   99.78
2009-02-10 09:40:00  86.63   99.75
2009-02-10 09:41:00  86.52   99.66

Data.info()

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 961276 entries, 2009-02-10 09:31:00 to 2016-02-29 19:59:00
Data columns (total 2 columns):
AAA    961276 non-null float64
BBB    961276 non-null float64
dtypes: float64(2)
memory usage: 22.0 MB

Data.index

Out[25]: 
DatetimeIndex(['2009-02-10 09:31:00', '2009-02-10 09:36:00',
               '2009-02-10 09:38:00', '2009-02-10 09:40:00',
               '2009-02-10 09:41:00', '2009-02-10 09:44:00',
               '2009-02-10 09:45:00', '2009-02-10 09:46:00',
               '2009-02-10 09:47:00', '2009-02-10 09:48:00',
               ...
               '2016-02-29 19:41:00', '2016-02-29 19:42:00',
               '2016-02-29 19:43:00', '2016-02-29 19:50:00',
               '2016-02-29 19:52:00', '2016-02-29 19:53:00',
               '2016-02-29 19:56:00', '2016-02-29 19:57:00',
               '2016-02-29 19:58:00', '2016-02-29 19:59:00'],
              dtype='datetime64[ns]', name='Time', length=961276, freq=None)

要对数据重新采样，我执行以下操作:

tframe = '60T'
hr_mean = Data.resample(tframe).mean()

作为输出，我得到了只有两个数字的 Pandas 系列:

In[26]: hr_mean
Out[26]: 
AAA    156.535198
BBB     30.197029
dtype: float64

如果我选择不同的时间范围或重采样函数，我会得到相同的行为。

Answer 1

您显示的行为是旧 Pandas 版本( Pandas < 0.18)的预期行为。较新的 pandas 版本更改了 resample API，您在此处看到其中一个棘手的案例。

在v0.18之前，resample使用how关键字指定如何重采样，直接返回一个重采样的帧/系列:

In [5]: data = pd.DataFrame(np.random.randn(180, 2), columns=['AAA', 'BBB'], index=pd.date_range("2016-06-01", periods=180, freq='1T'))

# how='mean' is the default, so this is the same as data.resample('60T')
In [6]: data.resample('60T', how='mean')  
Out[6]:
                          AAA       BBB
2016-06-01 00:00:00  0.100026  0.210722
2016-06-01 01:00:00  0.093662 -0.078066
2016-06-01 02:00:00 -0.114801  0.002615

# calling .mean() now calculates the mean of each column, resulting in the following series:
In [7]: data.resample('60T', how='mean').mean()
Out[7]:
AAA    0.026296
BBB    0.045090
dtype: float64

In [8]: pd.__version__
Out[8]: u'0.17.1'

从 0.18.0 开始，resample 本身是一个延迟操作，这意味着您首先必须调用一个方法(在本例中为 mean())来执行实际重采样:

In [4]: data.resample('60T')
Out[4]: DatetimeIndexResampler [freq=<60 * Minutes>, axis=0, closed=left, label=left, convention=start, base=0]

In [5]: data.resample('60T').mean()
Out[5]:
                          AAA       BBB
2016-06-01 00:00:00 -0.059038  0.102275
2016-06-01 01:00:00 -0.141429 -0.021342
2016-06-01 02:00:00 -0.073341 -0.150091

In [6]: data.resample('60T').mean().mean()
Out[6]:
AAA   -0.091270
BBB   -0.023052
dtype: float64

In [7]: pd.__version__
Out[7]: '0.18.1'

参见 http://pandas.pydata.org/pandas-docs/stable/whatsnew.html#resample-api用于解释 API 的变化。