Question

我有一个像这样的pandas数据框：

                              date  price  volume
0       2017-10-24 01:00:07.870000  51.90       1
1       2017-10-24 01:00:10.167000  51.90       1
2       2017-10-24 01:00:11.370000  51.89      -1
3       2017-10-24 01:00:11.370000  51.89      -6
4       2017-10-24 01:00:12.573000  51.90       5
5       2017-10-24 01:00:13.573000  51.89      -2
6       2017-10-24 01:00:13.776000  51.90       1
7       2017-10-24 01:00:21.276000  51.89      -1
8       2017-10-24 01:00:21.276000  51.88      -1
9       2017-10-24 01:00:21.276000  51.88      -2
10      2017-10-24 01:00:29.979000  51.89       1

如果我想将其转换为numpy数组并访问日期的分钟属性，则可能。

>>> array_df = df.values

>>> array_df[:,0] = np.array(array_df[:,0], dtype='datetime64[ms]')

>>> array_df
array([[datetime.datetime(2017, 10, 24, 1, 0, 7, 870000), 51.9, 1],
    [datetime.datetime(2017, 10, 24, 1, 0, 10, 167000), 51.9, 1],
    [datetime.datetime(2017, 10, 24, 1, 0, 11, 370000), 51.89, -1],
    ..., dtype=object)

>>> array_df[0][0].minute
0

但是当我创建一个具有相同datetime64[ms]类型的结构化数组时，我无法访问其分钟属性。

>>> array_structured = np.zeros(10, dtype=[('index', np.int32),
                                    ('date', 'datetime64[ms]'),
                                    ('price', np.float32),
                                    ('neg_value', np.int32),
                                    ('pos_value', np.int32)])

>>> array_structured
array([(0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0),
    (0, '1970-01-01T00:00:00.000',  0., 0, 0)],
    dtype=[('index', '<i4'), ('date', '<M8[ms]'), ('price', '<f4'), ('neg_value', '<i4'), ('pos_value', '<i4')])

>>> array_structured['date'][0] = np.datetime64('2017-10-24 01:00:07.870000')

>>> array_structured['date'][0].minute
Traceback (most recent call last):
File "<console>", line 1, in <module>
AttributeError: 'numpy.datetime64' object has no attribute 'minute'

即使他们的dtypes与datetime64[ms]相同，为什么date of array_structured在minute attribute拥有date of array_df时却没有{{1}}？

Answer 1

In [57]: data
Out[57]: 
array([('1971-01-01T00:00:00.000',  0.), ('1972-01-01T00:00:00.000',  0.),
       ('2017-10-31T00:00:00.000',  0.)],
      dtype=[('date', '<M8[ms]'), ('price', '<f4')])
In [58]: adate = data['date'][0]
In [59]: adate
Out[59]: numpy.datetime64('1971-01-01T00:00:00.000')

数组的元素没有minute之类的属性。但是当使用Python或item提取到tolist时，它们会成为“datetime个对象：

In [68]: data['date'].tolist()
Out[68]: 
[datetime.datetime(1971, 1, 1, 0, 0),
 datetime.datetime(1972, 1, 1, 0, 0),
 datetime.datetime(2017, 10, 31, 0, 0)]

In [61]: adate.item()
Out[61]: datetime.datetime(1971, 1, 1, 0, 0)
In [62]: adate.item().minute
Out[62]: 0
In [63]: adate.item().year
Out[63]: 1971
In [65]: [d.year for d in data['date'].tolist()]
Out[65]: [1971, 1972, 2017]

他们也可以使用'astype'转换为其他'时间单位'：

In [66]: data['date'].astype('datetime64[Y]')
Out[66]: array(['1971', '1972', '2017'], dtype='datetime64[Y]')
In [67]: data['date'].astype('datetime64[m]')
Out[67]: array(['1971-01-01T00:00', '1972-01-01T00:00', '2017-10-31T00:00'], dtype='datetime64[m]')

datetime64类型没有分钟属性吗？

1 个答案: