Numpy中的时间类型

从Numpy1.7开始，已经有了原生的日期-时间支持，基本类型称为datetime64。

In [1]: import numpy as np

In [2]: nd = np.datetime64('2018-05-08')

In [3]: nd
Out[3]: numpy.datetime64('2018-05-08')

和datetime一样，datetime64对象可以表示为字符串对象。

In [4]: np.datetime_as_string(nd)
Out[4]: '2018-05-08'

每个这类对象都保存在元数据，可以通过date_time_date方法访问，两个主要的组成部分是频率信息和单位（D:天）。

In [5]: np.datetime_data(nd)
Out[5]: ('D', 1)

 

从datetime对象中构造datetime64对象：

In [7]: d = datetime.datetime(2018,7,31,12,16,30,500000)

In [8]: d
Out[8]: datetime.datetime(2018, 7, 31, 12, 16, 30, 500000)

In [10]: ng = np.datetime64(d)

In [11]: ng
Out[11]: numpy.datetime64('2018-07-31T12:16:30.500000')

同样，可以将datetime64对象转换成datetime对象：

In [12]: nd.astype(datetime.datetime)
Out[12]: datetime.date(2018, 5, 8)

In [13]: nd.astype(datetime.date)
Out[13]: datetime.date(2018, 5, 8)

另外还有另外一种不常见的datetime64的构造方法：（提供一个字符串解和频率信息）

In [15]: nh = np.datetime64('2015-10','D')

In [16]: nh
Out[16]: numpy.datetime64('2015-10-01')

 

 

使用arange函数生成一个日期范围，当然可以指定日期范围

In [17]: np.array(['2015-10-01','2015-10-05','2015-10-10'],dtype='datetime64')
Out[17]: array(['2015-10-01', '2015-10-05', '2015-10-10'], dtype='datetime64[D]')

In [18]: np.arange('2018-10-01','2018-10-08',dtype='datetime64')
Out[18]:
array(['2018-10-01', '2018-10-02', '2018-10-03', '2018-10-04',
       '2018-10-05', '2018-10-06', '2018-10-07'], dtype='datetime64[D]')

In [19]: np.arange('2018-10-01','2018-10-08',dtype='datetime64[D]')
Out[19]:
array(['2018-10-01', '2018-10-02', '2018-10-03', '2018-10-04',
       '2018-10-05', '2018-10-06', '2018-10-07'], dtype='datetime64[D]')

In [20]: np.arange('2018-10-01','2018-10-31',dtype='datetime64[W]')
Out[20]:
array(['2018-09-27', '2018-10-04', '2018-10-11', '2018-10-18'],
      dtype='datetime64[W]')

In [21]: dt1 = np.arange('2018-01-01 10:10:10','2018-01-02 10:10:10',dtype='datetime64[h]')[:10]

In [22]: dt1
Out[22]:
array(['2018-01-01T10', '2018-01-01T11', '2018-01-01T12', '2018-01-01T13',
       '2018-01-01T14', '2018-01-01T15', '2018-01-01T16', '2018-01-01T17',
       '2018-01-01T18', '2018-01-01T19'], dtype='datetime64[h]')

In [36]: np.arange('2016-01-01T00:00:00','2016-01-02T00:00:00',dtype='datetime64[s]')[:20]
Out[36]:
array(['2016-01-01T00:00:00', '2016-01-01T00:00:01',
       '2016-01-01T00:00:02', '2016-01-01T00:00:03',
       '2016-01-01T00:00:04', '2016-01-01T00:00:05',
       '2016-01-01T00:00:06', '2016-01-01T00:00:07',
       '2016-01-01T00:00:08', '2016-01-01T00:00:09',
       '2016-01-01T00:00:10', '2016-01-01T00:00:11',
       '2016-01-01T00:00:12', '2016-01-01T00:00:13',
       '2016-01-01T00:00:14', '2016-01-01T00:00:15',
       '2016-01-01T00:00:16', '2016-01-01T00:00:17',
       '2016-01-01T00:00:18', '2016-01-01T00:00:19'],
      dtype='datetime64[s]')

In [37]: np.arange('2016-01-01T00:00:00','2016-01-02T00:00:00',dtype='datetime64[ms]')[:20]
Out[37]:
array(['2016-01-01T00:00:00.000', '2016-01-01T00:00:00.001',
       '2016-01-01T00:00:00.002', '2016-01-01T00:00:00.003',
       '2016-01-01T00:00:00.004', '2016-01-01T00:00:00.005',
       '2016-01-01T00:00:00.006', '2016-01-01T00:00:00.007',
       '2016-01-01T00:00:00.008', '2016-01-01T00:00:00.009',
       '2016-01-01T00:00:00.010', '2016-01-01T00:00:00.011',
       '2016-01-01T00:00:00.012', '2016-01-01T00:00:00.013',
       '2016-01-01T00:00:00.014', '2016-01-01T00:00:00.015',
       '2016-01-01T00:00:00.016', '2016-01-01T00:00:00.017',
       '2016-01-01T00:00:00.018', '2016-01-01T00:00:00.019'],
      dtype='datetime64[ms]')

简单应用：

import matplotlib.pyplot as plt
%matplotlib inline
np.random.seed(3000)
rnd = np.random.standard_normal(len(dt1)).cumsum()  ** 2
fig = plt.figure()
plt.plot(dt1.astype(datetime.datetime), rnd)
plt.grid(True)
fig.autofmt_xdate()

效果图：