關(guān)于時(shí)間的處理,，Python中自帶的處理時(shí)間的模塊就有time ,、datetime、calendar,，另外還有擴(kuò)展的第三方庫(kù),，如dateutil等等。通過(guò)這些途徑可以隨心所欲地用Python去處理時(shí)間,。當(dāng)我們用NumPy庫(kù)做數(shù)據(jù)分析時(shí),，如何轉(zhuǎn)換時(shí)間呢？

在NumPy 1.7版本開(kāi)始，它的核心數(shù)組（ndarray）對(duì)象支持datetime相關(guān)功能,，由于’datetime’這個(gè)數(shù)據(jù)類型名稱已經(jīng)在Python自帶的datetime模塊中使用了,， NumPy中時(shí)間數(shù)據(jù)的類型稱為’datetime64’。

單個(gè)時(shí)間格式字符串轉(zhuǎn)換為numpy的datetime對(duì)象,，可使用datetime64實(shí)例化一個(gè)對(duì)象,，如下所示：

#時(shí)間字符串轉(zhuǎn)numpy.datetime64
datetime_nd=np.datetime64('2019-01-01')
print(type(datetime_nd))#<class 'numpy.datetime64'>

反過(guò)來(lái)numpy的datetime對(duì)象轉(zhuǎn)換為時(shí)間格式字符串，可使用datetime_as_string()函數(shù),，如下所示：

#numpy.datetime64轉(zhuǎn)時(shí)間字符串
datetime_str=np.datetime_as_string(datetime_nd)
print(type(datetime_str))#<class 'numpy.str_'>

從時(shí)間格式字符串?dāng)?shù)組去創(chuàng)建numpy的datetime對(duì)象數(shù)組（array）時(shí),，可以直接使用numpy.array()函數(shù),，指定dtype為’datetime64’,，這樣的話數(shù)組中的元素為’datetime64’類型，如下所示：

datetime_array = np.array(['2019-01-05','2019-01-02','2019-01-03'], dtype='datetime64')
print(datetime_array)#['2019-01-05' '2019-01-02' '2019-01-03']
print(type(datetime_array))#<class 'numpy.ndarray'>
print(type(datetime_array[0]))#<class 'numpy.datetime64'>

也可以通過(guò)numpy.arange()函數(shù),，給定時(shí)間起始范圍去創(chuàng)建numpy的datetime對(duì)象數(shù)組（array）,，指定dtype為’datetime64’時(shí)默認(rèn)以日為時(shí)間間隔，如下所示：

datetime_array = np.arange('2019-01-05','2019-01-10', dtype='datetime64')
print(datetime_array)#['2019-01-05' '2019-01-06' '2019-01-07' '2019-01-08' '2019-01-09']

設(shè)定numpy.arange()函數(shù)中的dtype參數(shù),，可以調(diào)整時(shí)間的間隔,，比如以年、月,、周,，甚至小時(shí)、分鐘,、毫秒程度的間隔生成時(shí)間數(shù)組,，這點(diǎn)和Python的datetime模塊是一樣的，分為了date單位和time單位,。如下所示：

# generate year datetime array
datetime_array = np.arange('2018-01-01','2020-01-01', dtype='datetime64[Y]')
print(datetime_array)#['2018' '2019']
# generate month datetime array
datetime_array = np.arange('2019-01-01','2019-10-01', dtype='datetime64[M]')
print(datetime_array)#['2019-01' '2019-02' '2019-03' '2019-04' '2019-05' '2019-06' '2019-07' '2019-08' '2019-09']
# generate week datetime array
datetime_array = np.arange('2019-01-05','2019-02-10', dtype='datetime64[W]')
print(datetime_array)#['2019-01-03' '2019-01-10' '2019-01-17' '2019-01-24' '2019-01-31']
# generate ms datetime array
datetime_array = np.arange('2019-01-05','2019-01-10', dtype='datetime64[ms]')
print(datetime_array)
#['2019-01-05T00:00:00.000' '2019-01-05T00:00:00.001'
# '2019-01-05T00:00:00.002' ... '2019-01-09T23:59:59.997'
# '2019-01-09T23:59:59.998' '2019-01-09T23:59:59.999']

將numpy.datetime64轉(zhuǎn)化為datetime格式轉(zhuǎn)換為datetime格式,，可使用astype()方法轉(zhuǎn)換數(shù)據(jù)類型，如下所示：

#numpy.datetime64轉(zhuǎn)化為datetime格式
datetime_df=datetime_nd.astype(datetime.datetime)
print(type(datetime_df))#<class 'datetime.date'>

另外,，numpy也提供了datetime.timedelta類的功能,，支持兩個(gè)時(shí)間對(duì)象的運(yùn)算，得到一個(gè)時(shí)間單位形式的數(shù)值,。因?yàn)閚umpy的核心數(shù)組（ndarray）對(duì)象沒(méi)有物理量系統(tǒng)（physical quantities system）,，所以創(chuàng)建了timedelta64數(shù)據(jù)類型來(lái)補(bǔ)充datetime64。datetime和timedelta結(jié)合提供了更簡(jiǎn)單的datetime計(jì)算方法,。如下所示：

# numpy.datetime64 calculations
datetime_delta = np.datetime64('2009-01-01') - np.datetime64('2008-01-01')
print(datetime_delta)#366 days
print(type(datetime_delta))#<class 'numpy.timedelta64'>
datetime_delta = np.datetime64('2009') + np.timedelta64(20, 'D')
print(datetime_delta)#2009-01-21
datetime_delta = np.datetime64('2011-06-15T00:00') + np.timedelta64(12, 'h')
print(datetime_delta)#2011-06-15T12:00
datetime_delta = np.timedelta64(1,'W') / np.timedelta64(1,'D')
print(datetime_delta)#7.0