使用 origin='end_day' 重新采样

Question

我不明白origin='end_day' 做了什么。

docs 举个例子：

>>> start, end = '2000-10-01 23:30:00', '2000-10-02 00:30:00'
>>> rng = pd.date_range(start, end, freq='7min')
>>> ts = pd.Series(np.arange(len(rng)) * 3, index=rng)
>>> ts 
2000-10-01 23:30:00     0
2000-10-01 23:37:00     3
2000-10-01 23:44:00     6
2000-10-01 23:51:00     9
2000-10-01 23:58:00    12
2000-10-02 00:05:00    15
2000-10-02 00:12:00    18
2000-10-02 00:19:00    21
2000-10-02 00:26:00    24
Freq: 7T, dtype: int32
>>> ts.resample('17min', origin='end_day').sum()
2000-10-01 23:38:00     3
2000-10-01 23:55:00    15
2000-10-02 00:12:00    45
2000-10-02 00:29:00    45
Freq: 17T, dtype: int32

文档这样解释origin='end_day'：

‘end_day’：原点是最后一天的天花板午夜

据我所知，这条线

ts.resample('17min', origin='end_day').sum()

应该等价于

ts.resample('17min', origin=ts.index.max().ceil('1d')).sum()

但是，传递时间戳ts.index.max().ceil('1d') 会产生不同的结果：

>>> ts.resample('17min', origin=ts.index.max().ceil('1d')).sum() 
2000-10-01 23:21:00     3
2000-10-01 23:38:00    15
2000-10-01 23:55:00    27
2000-10-02 00:12:00    63

我正在寻找对这种差异的解释，也许是对 'end_day' 参数的一般描述比文档提供的更好。

编辑：我正在使用pandas 1.3.5

Answer 1

origin='end_day' 的真正等价物是：

>>> ts.resample('17min', origin=ts.index.max().ceil('D'), 
                closed='right', label='right').sum()

2000-10-01 23:38:00     3
2000-10-01 23:55:00    15
2000-10-02 00:12:00    45
2000-10-02 00:29:00    45
Freq: 17T, dtype: int64

更新 1：

1. 如果我使用 origin='end_day' 但还明确传入已关闭且标签不是“正确”，该怎么办？为此定义的行为在哪里？

来自resample的source code：

            # The backward resample sets ``closed`` to ``'right'`` by default
            # since the last value should be considered as the edge point for
            # the last bin. When origin in "end" or "end_day", the value for a
            # specific ``Timestamp`` index stands for the resample result from
            # the current ``Timestamp`` minus ``freq`` to the current
            # ``Timestamp`` with a right close.
            if origin in ["end", "end_day"]:
                if closed is None:
                    closed = "right"
                if label is None:
                    label = "right"
            else:
                if closed is None:
                    closed = "left"
                if label is None:
                    label = "left"

更新 2a：

1. 考虑df = pd.DataFrame(index=pd.date_range(start='2021-04-22 01:00:00', end='2021-04-28 01:00', freq='1d'), data=range(7))。现在 df.resample(rule='7d', origin='end_day') 崩溃并出现 ValueError。

如果您没有明确设置closed 参数，则将resample 设置为right，因为origin='end_day'（见上文）。所以origin 现在是 '2021-04-29' 并且第一个 bin 值是 '2021-04-22' 被排除在外。你有一种情况Values falls before first bin:

df = pd.DataFrame(index=pd.date_range(start='2021-04-22 01:00:00', end='2021-04-28 01:00', freq='1d'), data=range(7))
df.resample(rule='7d', origin='end_day', closed='left')  # <- HERE

更新 2b：

如果“2021-04-22”是第一个 bin，那么哪个时间戳不在其中？ '2021-04-22 01:00:00' 更晚了，对吧？

df = pd.DataFrame(index=pd.date_range(start='2021-04-21 01:00:00', end='2021-04-28 01:00', freq='1d'), data=range(8))
print(df)

# Output:
                     0
2021-04-21 01:00:00  0
2021-04-22 01:00:00  1
2021-04-23 01:00:00  2
2021-04-24 01:00:00  3
2021-04-25 01:00:00  4
2021-04-26 01:00:00  5
2021-04-27 01:00:00  6
2021-04-28 01:00:00  7

有了这个样本，我想你应该更清楚：

# closed='right' (default)
>>> df.resample(rule='7d', origin='end_day').sum()
             0
2021-04-22   1  # ('2021-04-15', '2021-04-22']
2021-04-29  27  # ('2021-04-22', '2021-04-29']

# closed='left'
>>> df.resample(rule='7d', origin='end_day', closed='left').sum()
             0
2021-04-22   0  # ['2021-04-15', '2021-04-22')
2021-04-29  28  # ['2021-04-22', '2021-04-29')

bin_edges

bin_edges 的值为：

# closed='right' (default)
>>> bin_edges
[1618531199999999999 1619135999999999999 1619740799999999999]

# after conversion
DatetimeIndex(['2021-04-15 23:59:59.999999999',
               '2021-04-22 23:59:59.999999999',
               '2021-04-29 23:59:59.999999999'],
              dtype='datetime64[ns]', freq=None)


# closed='left'
>>> bin_edges
[1618444800000000000 1619049600000000000 1619654400000000000]

# after conversion
DatetimeIndex(['2021-04-15',
               '2021-04-22',
               '2021-04-29'],
              dtype='datetime64[ns]', freq=None)