Source code for spacepy.time

#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""Time conversion, manipulation and implementation of Ticktock class

Notes
=====
The handling of time, in particular the conversions between representations,
can be more complicated than it seems on the surface. This can result in
some surprising behavior, particularly when requiring second-level accuracy and
converting between time systems outside of the period 1972 to present.
It is strongly recommended to use TAI if transferring times between SpacePy
and other libraries. TAI has a consistent, unambiguous definition and no
discontinuities.

Some time systems (e.g. the UTC representation via datetime) cannot represent
times during a leapsecond. SpacePy represents all these times as the latest
representable time in the day, e.g.::

    >>> spacepy.time.Ticktock('2008-12-31T23:59:60').UTC[0]
    datetime.datetime(2008, 12, 31, 23, 59, 59, 999999)

Conversions between continuous time representations (e.g. TAI), leap second
aware representations (e.g. ISO timestrings), and those that ignore leap
seconds (e.g. UTC datetime, Unix time) are well-defined between the
introduction of the leap second system to UTC in 1972 and the present.
For systems that cannot represent leap seconds, the leap second moment is
considered not to exist. For example, from 23:59:59 on 2008-12-31 to 00:00:00
on 2009-01-01 is two seconds, but only represents a one-second increment in
Unix time. Details are also discussed in the individual time representations.

UTC times more than six months in the future are not well-defined, since
the schedule of leap second insertion is not known in advance. SpacePy
performs conversions assuming there are no leapseconds after those which have
been announced by IERS.

Between 1960 and 1972, UTC was defined by means of fractional leap
seconds and a varying-length second. From 1958 (when UTC was set equal
to TAI) and 1972, SpacePy treats UTC time similar to after 1972, with
a consistent second the same length of the SI second, and applying a
full leap second before the beginning of January and July if UTC - UT1
exceeded 0.4s. The difference with other methods of calculating UTC is
less than half a second.

.. versionchanged:: 0.2.3
   The application of post-1972 rules to 1958-1927 is new in
   0.2.3. Before, SpacePy applied leap seconds wherever there was an
   entry in the USNO record of TAI-UTC, rounding fractional total leap
   second counts to the integer (0.5 rounds up). The UTC second was still
   treated as the same length as the SI second (i.e., rate changed were
   not applied.) This resulted in the application of six leap seconds at
   the beginning of 1972. The discrepancy with other means of calculating
   TAI-UTC was as much as five seconds by the end of this period.

.. versionchanged:: 0.2.2
   Before 0.2.2, SpacePy truncated fractional leapseconds rather than rounding.

Before 1958, UTC is not defined. SpacePy assumes days of constant length
86400 seconds, equal to the SI second. This is almost guaranteed to be wrong;
for times well out of the space era, it is strongly recommended to work
consistently in either a continuous time system (e.g. TAI) or a day-based
system (e.g. JD).

SpacePy assumes dates including and after 1582-10-15 to be in the Gregorian
calendar and dates including and before 1582-10-04 to be Julian. 10-05 through
10-14 do not exist. This change is ignored for continuously-running non leap
second aware timebases: CDF and RDT.

See the :class:`Ticktock` documentation and its various ``get`` functions for
more details on the exact definitions of time systems used by SpacePy.

Examples:
=========

>>> import spacepy.time as spt
>>> import datetime as dt

Day of year calculations

>>> dts = spt.doy2date([2002]*4, range(186,190), dtobj=True)
>>> dts
[datetime.datetime(2002, 7, 5, 0, 0),
datetime.datetime(2002, 7, 6, 0, 0),
datetime.datetime(2002, 7, 7, 0, 0),
datetime.datetime(2002, 7, 8, 0, 0)]

>>> dts = spt.Ticktock(dts,'UTC')
>>> dts.DOY
array([ 186.,  187.,  188.,  189.])

Ticktock object creation

>>> isodates = ['2009-12-01T12:00:00', '2009-12-04T00:00:00', '2009-12-06T12:00:00']
>>> dts = spt.Ticktock(isodates, 'ISO')

OR

>>> dtdates = [dt.datetime(2009,12,1,12), dt.datetime(2009,12,4), dt.datetime(2009,12,6,12)]
>>> dts = spt.Ticktock(dtdates, 'UTC')

ISO time formatting

>>> dts = spt.tickrange('2009-12-01T12:00:00','2009-12-06T12:00:00',2.5)

OR

>>> dts = spt.tickrange(dt.datetime(2009,12,1,12),dt.datetime(2009,12,6,12), \
    dt.timedelta(days=2, hours=12))

>>> dts
Ticktock( ['2009-12-01T12:00:00', '2009-12-04T00:00:00', '2009-12-06T12:00:00'] ), dtype=ISO

>>> dts.isoformat()
Current ISO output format is %Y-%m-%dT%H:%M:%S
Options are: [('seconds', '%Y-%m-%dT%H:%M:%S'), ('microseconds', '%Y-%m-%dT%H:%M:%S.%f')]

>>> dts.isoformat('microseconds')
>>> dts.ISO
['2009-12-01T12:00:00.000000',
 '2009-12-04T00:00:00.000000',
 '2009-12-06T12:00:00.000000']

Time manipulation

>>> new_dts = dts + tdelt
>>> new_dts.UTC
[datetime.datetime(2009, 12, 2, 18, 0),
 datetime.datetime(2009, 12, 5, 6, 0),
 datetime.datetime(2009, 12, 7, 18, 0)]

Other time formats

>>> dts.RDT  # Gregorian ordinal time
array([ 733742.5,  733745. ,  733747.5])

>>> dts.GPS # GPS time
array([  9.43704015e+08,   9.43920015e+08,   9.44136015e+08])

>>> dts.JD # Julian day
array([ 2455167. ,  2455169.5,  2455172. ])

And so on.

.. currentmodule:: spacepy.time

.. NOTE... there is an error with this reference

Authors: Steve Morley, Josef Koller, Brian Larsen, Jon Niehof
Institution: Los Alamos National Laboratory
Contact: smorley@lanl.gov,


Copyright 2010 Los Alamos National Security, LLC.

"""
from __future__ import absolute_import

import bisect
try:
    from collections.abc import Callable, MutableSequence
except ImportError:
    from collections import Callable, MutableSequence
import datetime

try:
    from itertools import izip as zip
except ImportError:
    pass  # just use system zip. In python3 itertools.izip is just python zip
import os.path
import re
import time
import warnings

try:
    import astropy.time
    HAVE_ASTROPY = True
except ImportError:
    HAVE_ASTROPY = False
import dateutil.parser as dup
import numpy as np

import spacepy
from spacepy import help
import spacepy.datamodel

__contact__ = 'Steve Morley, smorley@lanl.gov'


# -----------------------------------------------
# Ticktock class
# -----------------------------------------------
[docs]class Ticktock(MutableSequence): """ Ticktock( data, dtype ) Ticktock class holding various time coordinate systems (TAI, UTC, ISO, JD, MJD, GPS, UNX, RDT, CDF, DOY, eDOY, APT) Possible input data types: ISO ISO standard format like '2002-02-25T12:20:30' UTC datetime object with UTC time TAI Elapsed seconds since 1958-1-1 (includes leap seconds) GPS Elapsed seconds since 1980-1-6 (includes leap seconds) UNX Elapsed seconds since 1970-1-1 ignoring leapseconds (all days have 86400 secs). JD Julian days elapsed MJD Modified Julian days RDT Rata Die days elapsed since 0001-1-1 CDF CDF Epoch type: float milliseconds since 0000-1-1 ignoring leapseconds APT AstroPy :class:`~astropy.time.Time`. Requires AstroPy 1.0. (New in version 0.2.2.) Possible output data types: All those listed above, plus: DOY Integer day of year, starts with day 1 eDOY Fractional day of year, starts at day 0 It is strongly recommended to access various time systems via the attributes listed above, as in the examples. They will be calculated automatically if necessary. Using the ``get`` methods will force a recalculation. The original input data will always be available as the ``data`` attribute. .. versionchanged:: 0.2.2 In earlier versions of SpacePy, most values were derived from the ``datetime``-based ``UTC`` representation. This did not properly handle leap seconds in many cases. Now most are derived from ``TAI`` (exceptions being ``DOY`` and ``eDOY``). In addition to differences around actual leap seconds, this may result in small differences between versions of SpacePy, with relative magnitude on the order of the resolution of a 64-bit float (2e-16). For times in the modern era, this is about 50 microseconds (us) for ``JD``, 15 us for ``CDF``, 1.5 us for ``RDT``, 1 us for ``MJD``, and 360 *nanoseconds* for ``TAI``. The relationships between parameters and how they are calculated are listed in the ``get`` methods and illustrated below. .. only:: latex .. image:: ../images/ticktock_relationships.pdf .. only:: html .. image:: ../images/ticktock_relationships.svg :target: ../_images/ticktock_relationships.svg Parameters ========== data : array_like (int, datetime, float, string) time stamp dtype : string {`CDF`, `ISO`, `UTC`, `TAI`, 'GPS', `UNX`, `JD`, `MJD`, `RDT`, `APT`} or function data type for data, if a function it must convert input time format to Python datetime Returns ======= out : Ticktock instance with self.data, self.dtype, self.UTC etc Notes ===== UTC data type is implemented as Python datetime, which cannot represent leap seconds. The time within a leap second is regarded as not happening. The CDF data type is the older CDF_EPOCH time type, not the newer CDF_TIME_TT2000. It similarly cannot represent leap seconds. Year 0 is considered a leap year. Other Parameters ================ isoformat : str, optional .. versionadded:: 0.2.2 Format string used for parsing and outputting ISO format. Input is not forced to be in this format; it is tried first, and other common formats tried if parsing fails. Because of this, if ISO input is in a consistent format, specifying this can speed up the input parsing. Can be changed on existing ``Ticktock`` with :meth:`isoformat` method. Default ``'%Y-%m-%dT%H:%M:%S'``. See Also ======== datetime.datetime.strptime, isoformat Examples ======== >>> x = Ticktock([2452331.0142361112, 2452332.0142361112], 'JD') >>> x.ISO dmarray(['2002-02-25T12:20:30', '2002-02-26T12:20:30'], dtype='|S19') >>> x.DOY # Day of year dmarray([ 56., 57.]) >>> y = Ticktock(['01-01-2013', '20-03-2013'], lambda x: datetime.datetime.strptime(x, '%d-%m-%Y')) >>> y.UTC dmarray([2013-01-01 00:00:00, 2013-03-20 00:00:00], dtype=object) >>> y.DOY # Day of year dmarray([ 1., 79.]) .. autosummary:: ~Ticktock.append ~Ticktock.argsort ~Ticktock.convert ~Ticktock.getAPT ~Ticktock.getCDF ~Ticktock.getDOY ~Ticktock.getGPS ~Ticktock.getISO ~Ticktock.getJD ~Ticktock.getMJD ~Ticktock.getRDT ~Ticktock.getTAI ~Ticktock.getUNX ~Ticktock.getUTC ~Ticktock.geteDOY ~Ticktock.getleapsecs ~Ticktock.isoformat ~Ticktock.now ~Ticktock.sort ~Ticktock.today ~Ticktock.update_items .. automethod:: append .. automethod:: argsort .. automethod:: convert .. automethod:: getAPT .. automethod:: getCDF .. automethod:: getDOY .. automethod:: getGPS .. automethod:: getISO .. automethod:: getJD .. automethod:: getMJD .. automethod:: getRDT .. automethod:: getTAI .. automethod:: getUNX .. automethod:: getUTC .. automethod:: geteDOY .. automethod:: getleapsecs .. automethod:: isoformat .. automethod:: now .. automethod:: sort .. automethod:: today .. automethod:: update_items """ _keylist = ['UTC', 'TAI', 'ISO', 'JD', 'MJD', 'UNX', 'RDT', 'CDF', 'GPS', 'DOY', 'eDOY', 'leaps'] if HAVE_ASTROPY: _keylist.append('APT') _keylist_upper = [key.upper() for key in _keylist] _isoformatstr = {'seconds': '%Y-%m-%dT%H:%M:%S', 'microseconds': '%Y-%m-%dT%H:%M:%S.%f'} def __init__(self, data, dtype=None, isoformat=None): self._isofmt = isoformat or self._isoformatstr['seconds'] if isinstance(data, Ticktock): dtype = data.data.attrs['dtype'] self.data = data.data else: try: spacepy.datamodel.dmarray(data)[0] except IndexError: self.data = spacepy.datamodel.dmarray([data]) else: self.data = spacepy.datamodel.dmarray(data) if not isinstance(dtype, Callable): if isinstance(self.data[0], (str, bytes)): dtype = 'ISO' elif str is bytes and isinstance(self.data[0], unicode): #Py2k dtype = 'ISO' elif isinstance(self.data[0], datetime.datetime): dtype = 'UTC' elif HAVE_ASTROPY and isinstance(self.data[0], astropy.time.Time): dtype = 'APT' # Recover original input (not dmarray), add axis if scalar self.data = astropy.time.Time([data]) if data.shape == ()\ else data elif self.data[0] > 1e13: dtype = 'CDF' elif dtype is None: raise ValueError('Unable to guess dtype from data; ' 'please specify dtype.') if dtype.upper() not in Ticktock._keylist_upper: raise ValueError("data type " + dtype + " not provided, only " + str(Ticktock._keylist)) else: # process input data using callable dtype to convert to datetime/UTC dtype_func = np.vectorize(dtype) self.data = dtype_func(self.data) self.UTC = no_tzinfo(self.data) # AstroPy time objects don't have attrs, but will accept one. if not hasattr(self.data, 'attrs'): self.data.attrs = {} try: self.data.attrs['dtype'] = dtype.upper() except AttributeError: self.data.attrs['dtype'] = str(dtype_func) else: # ISO is populated by update_items self.update_items('data') if dtype.upper() == 'TAI': self.TAI = self.data elif dtype.upper() == 'JD': self.JD = self.data elif dtype.upper() == 'MJD': self.MJD = self.data elif dtype.upper() == 'UNX': self.UNX = self.data elif dtype.upper() == 'RDT': self.RDT = self.data elif dtype.upper() == 'CDF': self.CDF = self.data elif dtype.upper() == 'UTC': self.UTC = no_tzinfo(self.data) elif dtype.upper() == 'APT': self.APT = self.data ## Brian and Steve were looking at this to see about making plot work directly on the object ## is also making iterate as an array of datetimes # def __iter__(self): # i = 0 # try: # while True: # v = self[i].UTC[0] # yield v # i += 1 # except IndexError: # return # ----------------------------------------------- def __str__(self): """ a.__str__() or a Will be called when printing Ticktock instance a Returns ======= out : string string representaion of the class Examples ======== >>> a = Ticktock('2002-02-02T12:00:03', 'ISO') >>> a Ticktock( ['2002-02-02T12:00:00'], dtype=ISO) """ return 'Ticktock( ' + str(self.data) + ', dtype=' + str(self.data.attrs['dtype'] + ')') __repr__ = __str__ # ----------------------------------------------- def __getstate__(self): """ Is called when pickling See Also http://docs.python.org/library/pickle.html """ odict = self.__dict__.copy() # copy the dict since we change it return odict def __setstate__(self, dict): """ Is called when unpickling See Also http://docs.python.org/library/pickle.html """ self.__dict__.update(dict) return # ----------------------------------------------- def __getitem__(self, idx): """ a.__getitem__(idx) or a[idx] Will be called when requesting items in this instance Parameters ========== idx : int the item index to get Returns ======= out : Ticktock Ticktock instance with requested values Examples ======== >>> a = Ticktock('2002-02-02T12:00:03', 'ISO') >>> a[0] '2002-02-02T12:00:00' See Also ======== a.__setitem__ """ return Ticktock(self.UTC[idx]) # ----------------------------------------------- def __setitem__(self, idx, vals): """ a.__setitem__(idx, vals) or a[idx] = vals Will be called setting items in this instance Parameters ========== idx : int integer numbers as index vals: {float, string, datetime} new values Examples ======== >>> a = Ticktock('2002-02-02T12:00:03', 'ISO') >>> a[0] = '2003-03-03T00:00:00' See Also ======== a.__getitem__ """ tmp = Ticktock(vals) if len(tmp) > 1: self.data[idx] = tmp.__getattribute__(self.data.attrs['dtype'])[:] else: self.data[idx] = tmp.__getattribute__(self.data.attrs['dtype'])[0] self.update_items('data') # ----------------------------------------------- def __delitem__(self, idx): """ a.__delitem(index) will be called when deleting items in the sequence """ self.data = np.delete(self.data, idx) self.update_items('data') # ----------------------------------------------- def __len__(self): """ a.__len__() or len(a) Will be called when requesting the length, i.e. number of items Returns ======= out : int length Examples ======== >>> a = Ticktock('2002-02-02T12:00:03', 'ISO') >>> a.len 1 """ return len(self.data) # ----------------------------------------------- def __gt__(self, other): if isinstance(other, datetime.datetime): other = Ticktock(other, 'UTC') return self.UNX > other.UNX def __lt__(self, other): if isinstance(other, datetime.datetime): other = Ticktock(other, 'UTC') return self.UNX < other.UNX def __ge__(self, other): if isinstance(other, datetime.datetime): other = Ticktock(other, 'UTC') return self.UNX >= other.UNX def __le__(self, other): if isinstance(other, datetime.datetime): other = Ticktock(other, 'UTC') return self.UNX <= other.UNX def __eq__(self, other): if isinstance(other, datetime.datetime): other = Ticktock(other, 'UTC') return self.UNX == other.UNX def __ne__(self, other): if isinstance(other, datetime.datetime): other = Ticktock(other, 'UTC') return self.UNX != other.UNX # ----------------------------------------------- def __sub__(self, other): """ a.__sub__(other) Will be called if a timedelta object is subtracted from this instance and returns a new Ticktock instance. If a Ticktock is subtracted from another Ticktock then a list of timedeltas is returned. Parameters ========== other : Ticktock or datetime.timedelta instance for comparison Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> dt = datetime.timedelta(3) >>> a - dt Ticktock( ['2002-02-05T12:00:00'] ), dtype=ISO See Also ======== __add__ """ if isinstance(other, datetime.timedelta): newobj = Ticktock(self.UTC - other, 'UTC') elif isinstance(other, Ticktock): if not (len(other) == len(self.data)) and not (len(other) == 1): raise ValueError('Ticktock lengths are mismatched, subtraction is not possible') same = True if len(other) == 1: same = False if same: return [datetime.timedelta(seconds=t - other.TAI[i]) for i, t in enumerate(self.TAI)] else: return [datetime.timedelta(seconds=t - other.TAI[0]) for t in self.TAI] elif hasattr(other, '__iter__'): if not isinstance(other[0], datetime.timedelta): raise TypeError("Data supplied for addition is of the wrong type") if not (len(other) == len(self.data)) or (len(other) == 1): raise TypeError("Data supplied for addition is of the wrong shape") same = True if len(other) == 1: same = False if same: newUTC = [utc - o for utc, o in zip(self.UTC, other)] else: newUTC = [utc - other for utc in self.UTC] newobj = Ticktock(newUTC, 'UTC') else: raise TypeError("unsupported operand type(s) for -: {0} and {1}".format(type(other), type(self))) return newobj # ----------------------------------------------- def __add__(self, other): """ a.__add__(other) Will be called if an iterable of datetime.timedeltas is added to this instance and returns a new Ticktock instance Parameters ========== other : datetime.timedelta instance for comparison Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> import datetime as dt >>> delt = dt.timedelta(minutes=1) >>> a + delt Ticktock( ['2002-02-02T12:01:00'] , dtype=ISO) See Also ======== __sub__ """ if isinstance(other, datetime.timedelta): newobj = Ticktock(self.UTC + other, 'UTC') elif hasattr(other, '__iter__'): if not isinstance(other[0], datetime.timedelta): raise TypeError("Data supplied for addition is of the wrong type") if not (len(other) == len(self.data)) or (len(other) == 1): raise TypeError("Data supplied for addition is of the wrong shape") same = True if len(other) == 1: same = False if same: newUTC = [utc + o for utc, o in zip(self.UTC, other)] else: newUTC = [utc + other for utc in self.UTC] newobj = Ticktock(newUTC, 'UTC') else: raise TypeError("unsupported operand type(s) for +: {0} and {1}".format(type(other), type(self))) return newobj def __radd__(self, other): """ a.__radd__(other) reverse add -- Will be called if this object is added to a datetime.timedelta and returns a new Ticktock instance Parameters ========== other : datetime.timedelta instance for comparison Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> import datetime as dt >>> delt = dt.timedelta(minutes=1) >>> a + delt Ticktock( ['2002-02-02T12:01:00'] , dtype=ISO) See Also ======== __sub__ """ return self.__add__(other) # ----------------------------------------------- def __getattr__(self, name): """ a.__getattr__(name) Will be called if attribute "name" is not found in Ticktock class instance. It will add TAI, RDT, etc Parameters ========== name : string a string from the list of time systems 'UTC', 'TAI', 'ISO', 'JD', 'MJD', 'UNX', 'RDT', 'CDF', 'DOY', 'eDOY', 'leaps' Returns ======= out: list, array requested values as either list/numpy array """ if name not in Ticktock._keylist: raise AttributeError("data type {} not provided, only {}".format( str(name), str(Ticktock._keylist))) if name.upper() == 'TAI': self.TAI = self.getTAI() if name.upper() == 'UTC': self.UTC = self.getUTC() if name.upper() == 'ISO': self.ISO = self.getISO() if name.upper() == 'JD': self.JD = self.getJD() if name.upper() == 'MJD': self.MJD = self.getMJD() if name.upper() == 'UNX': self.UNX = self.getUNX() if name.upper() == 'RDT': self.RDT = self.getRDT() if name.upper() == 'CDF': self.CDF = self.getCDF() if name.upper() == 'DOY': self.DOY = self.getDOY() if name.upper() == 'EDOY': self.eDOY = self.geteDOY() if name.upper() == 'GPS': self.GPS = self.getGPS() if name.upper() == 'APT': self.APT = self.getAPT() # if name == 'isoformat': self.__isofmt = self.isoformat() if name == 'leaps': self.leaps = self.getleapsecs() return getattr(self, name) # ----------------------------------------------- def insert(self, idx, val, dtype=None): """ insert values into the TickTock object .. note:: If more than one value to insert a slice must be specified as the index. See numpy.insert Parameters ========== idx : int, slice or sequence of ints Object that defines the index or indices before which `val` is inserted. val : array_like values to insert dtype : str (optional) must be specified if not CDF, ISO, or UTC """ fmt = self.data.attrs['dtype'] if not dtype: dum = Ticktock(val) else: dum = Ticktock(val, dtype=dtype) ival = getattr(dum, fmt) self.data = np.insert(self.data, idx, ival) self.update_items('data') # ----------------------------------------------- def remove(self, idx): """ a.remove(idx) This will remove the Ticktock value at index idx """ del self[idx] # -----------------------------------------------
[docs] def sort(self, kind='quicksort'): """ a.sort() This will sort the Ticktock values in place based on the values in `data`. If you need a stable sort use kind='mergesort' Other Parameters ================ kind : str Sort algorithm to use, default 'quicksort'. See Also ======== argsort, numpy.argsort """ idx = self.argsort(kind=kind) self.data = self.data[idx] self.update_items('data')
# -----------------------------------------------
[docs] def argsort(self, kind='quicksort'): """ idx = a.argsort() This will return the indices that would sort the Ticktock values Returns ======= out : list indices that would sort the Ticktock values Other Parameters ================ kind : str, optional Sort algorithm to use, default 'quicksort'. .. versionchanged:: 0.2.2 Default is now 'quicksort' to match numpy default; previously was 'mergesort'. See Also ======== argsort, numpy.argsort """ return np.argsort(self.TAI, kind=kind)
# -----------------------------------------------
[docs] def isoformat(self, fmt=None): """ a.update_items(b, attrib) This changes the self._isofmt attribute by and subsequently this function will update the ISO attribute. Parameters ========== fmt : string, optional """ if fmt is None: print('Current ISO output format is %s' % self._isofmt) print('Options are: {0}'.format([(k, Ticktock._isoformatstr[k]) for k in list(Ticktock._isoformatstr.keys())])) else: try: self._isofmt = Ticktock._isoformatstr[fmt] self.update_items('data') except KeyError: raise (ValueError('Not a valid option: Use {0}'.format(list(Ticktock._isoformatstr.keys()))))
# -----------------------------------------------
[docs] def update_items(self, attrib, cls=None): """ a.update_items(attrib) After changing the self.data attribute by either __setitem__ or __add__ etc this function will update all other attributes. This function is called automatically in __add__, __init__, and __setitem__. Parameters ========== attrib : str attribute that was updated; update others from this Other Parameters ================ cls : type .. deprecated:: 0.2.2 Class is now taken from the ``self`` of the bound instance. Class to use for finding possible attributes; now ignored. See Also ======== spacepy.Ticktock.__setitem__ spacepy.Ticktock.__add__ spacepy.Ticktock.__sub__ """ keylist = dir(self) # Formerly took arguments (cls, attrib) but there's nothing from # the class we can't get from the instance, so removed cls. # If we got two position arguments, though, that indicates the cls if cls is not None: attrib = cls warnings.warn( 'cls argument of update_items was deprecated in 0.2.2' ' and will be ignored.', DeprecationWarning) keylist.remove('data') if attrib != 'data': keylist.remove(attrib) attrib = attrib.upper() if attrib != self.data.attrs['dtype']: # Repopulating based on a different dtype, so make a temp # Ticktock to do the conversion. cls = type(self) dt = self.data.attrs['dtype'] self.data = getattr( cls(getattr(self, attrib), dtype=attrib), dt) if self.data.attrs['dtype'] in ( 'TAI', 'GPS', 'JD', 'MJD', 'RDT', 'CDF', 'UNX', 'ISO', 'APT'): if self.data.attrs['dtype'] == 'ISO': if 'UTC' in keylist: del self.UTC # Force recalc of UTC in TAI calc if 'ISO' in keylist: del self.ISO # Also will recalc the ISO del keylist[keylist.index('ISO')] # So no need to calc again self.TAI = self.getTAI() if 'UTC' in keylist and self.data.attrs['dtype'] != 'ISO': self.UTC = self.getUTC() else: self.UTC = self.getUTC() if 'TAI' in keylist: self.TAI = self.getTAI() for key in keylist: if key.upper() == 'ISO': self.ISO = self.getISO() if key.upper() == 'JD': self.JD = self.getJD() if key.upper() == 'MJD': self.MJD = self.getMJD() if key.upper() == 'UNX': self.UNX = self.getUNX() if key.upper() == 'RDT': self.RDT = self.getRDT() if key.upper() == 'CDF': self.CDF = self.getCDF() if key.upper() == 'DOY': self.DOY = self.getDOY() if key.upper() == 'EDOY': self.eDOY = self.geteDOY() if key.upper() == 'GPS': self.GPS = self.getGPS() if key.upper() == 'APT': self.APT = self.getAPT() if key == 'leaps': self.leaps = self.getleapsecs() return
# -----------------------------------------------
[docs] def convert(self, dtype): """ a.convert(dtype) convert a Ticktock instance into a new time coordinate system provided in dtype Parameters ========== dtype : string data type for new system, possible values are {`CDF`, `ISO`, `UTC`, `TAI`, `UNX`, `JD`, `MJD`, `RDT`} Returns ======= out : Ticktock Ticktock instance with new time coordinates Examples ======== >>> a = Ticktock(['2002-02-02T12:00:00', '2002-02-02T12:00:00'], 'ISO') >>> s = a.convert('TAI') >>> type(s) <class 'time.Ticktock'> >>> s Ticktock( [1391342432 1391342432] ), dtype=TAI See Also ======== CDF ISO UTC """ newdat = getattr(self, dtype) return Ticktock(newdat, dtype)
# -----------------------------------------------
[docs] def append(self, other): """ a.append(other) Will be called when another Ticktock instance has to be appended to the current one Parameters ========== other : Ticktock other (Ticktock instance) """ otherdata = getattr(other, self.data.attrs['dtype']) return Ticktock(np.append(self.data, otherdata), dtype=self.data.attrs['dtype'])
# -----------------------------------------------
[docs] def getCDF(self): """ a.getCDF() or a.CDF Return CDF Epoch time which is milliseconds since 0000-1-1 at 00:00:00.000. "Year zero" is a convention chosen by NSSDC to measure epoch values. This date is more commonly referred to as 1 BC and is considered a leap year. The CDF date/time calculations do not take into account the change to the Gregorian calendar or leap seconds, and cannot be directly converted into Julian date/times. Returns ``data`` if it was provided in CDF; otherwise always recalculates from the current value of ``TAI``, which will be created if necessary. Updates the ``CDF`` attribute. Returns ======= out : numpy array milliseconds since 0000-01-01T00:00:00 assuming no discontinuities. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.CDF array([ 6.31798704e+13]) See Also ======== getUTC getUNX getRDT getJD getMJD getISO getTAI getDOY geteDOY getAPT """ if self.data.attrs['dtype'] == 'CDF': # This should be the case from the constructor self.CDF = self.data return self.CDF CDFofTAI0 = 61788528000000. naive_tai = _tai_real_to_naive(self.TAI) cdf = naive_tai * 1e3 + CDFofTAI0 self.CDF = spacepy.datamodel.dmarray(cdf, attrs={'dtype': 'CDF'}) return self.CDF
# -----------------------------------------------
[docs] def getDOY(self): """ a.DOY or a.getDOY() extract DOY (days since January 1st of given year) Always recalculates from the current value of ``UTC``, which will be created if necessary. Updates the ``DOY`` attribute. Returns ======= out : numpy array day of the year Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.DOY array([ 33]) See Also ======== getUTC getUNX getRDT getJD getMJD getISO getTAI getDOY geteDOY getAPT """ DOY = [utc.toordinal() - datetime.date(utc.year, 1, 1).toordinal() + 1 for utc in self.UTC] self.DOY = spacepy.datamodel.dmarray(DOY, attrs={'dtype': 'DOY'}).astype(int) return self.DOY
# -----------------------------------------------
[docs] def geteDOY(self): """ a.eDOY or a.geteDOY() extract eDOY (elapsed days since midnight January 1st of given year) Always recalculates from the current value of ``UTC``, which will be created if necessary. Updates the ``eDOY`` attribute. Returns ======= out : numpy array days elapsed since midnight bbedJan. 1st Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.eDOY array([ 32.5]) See Also ======== getUTC getUNX getRDT getJD getMJD getISO getTAI getDOY geteDOY getAPT """ eDOY = [utc.toordinal() - datetime.date(utc.year, 1, 1).toordinal() for utc in self.UTC] eDOY = [edoy + utc.hour / 24. + utc.minute / 1440. + utc.second / 86400. + utc.microsecond / 86400000000. for edoy, utc in zip(eDOY, self.UTC)] self.eDOY = spacepy.datamodel.dmarray(eDOY, attrs={'dtype': 'eDOY'}) return self.eDOY
# -----------------------------------------------
[docs] def getJD(self): """ a.JD or a.getJD() convert dtype data into Julian Date (JD) Returns ``data`` if it was provided in JD; otherwise always recalculates from the current value of ``TAI``, which will be created if necessary. Updates the ``JD`` attribute. Returns ======= out : numpy array elapsed days since 4713 BCE 01-01T12:00 Notes ===== This is based on the UTC day, defined as JD(UTC), per the recommendation of `IAU General Assembly XXIII resolution B1 <https://www.iers.org/IERS/EN/Science/Recommendations/ resolutionB1.html>`_. Julian days with leapseconds are 86401 seconds long and each second is a smaller fraction of the day. Note this "stretching" is across the *Julian* Day, noon to noon. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.JD array([ 2452308.]) See Also ======== getUTC getUNX getRDT getJD getMJD getISO getTAI getDOY geteDOY getAPT """ if self.data.attrs['dtype'] == 'JD': # This should be the case from the constructor self.JD = self.data return self.JD # 1958-01-01T12:00 is JD 2436205.0 JDofTAI0 = 2436205.0 # Days-since-1958 is relative to noon self.JD = _days1958(self.TAI, leaps='rubber') + JDofTAI0 return self.JD
# -----------------------------------------------
[docs] def getMJD(self): """ a.MJD or a.getMJD() convert dtype data into MJD (modified Julian date) Returns ``data`` if it was provided in MJD; otherwise always recalculates from the current value of ``TAI`` which will be created if necessary. Updates the ``MJD`` attribute. Returns ======= out : numpy array elapsed days since 1858-11-17T00:00 (Julian date of 1858-11-17T12:00 was 2 400 000) Notes ===== This is based on the UTC day, defined as JD(UTC) - 2 400 000.5, per the recommendation of `IAU General Assembly XXIII resolution B1 <https://www.iers.org/IERS/EN/Science/Recommendations/ resolutionB1.html>`_. Julian days with leapseconds are 86401 seconds long and each second is a smaller fraction of the day. Note this "stretching" is across the *Julian* Day not the MJD, so it will affect the last half of the MJD before the leap second and the first half of the following MJD, so that MJD is always JD - 2 400 000.5 This also means that the MJD following a leap second does not begin exactly at midnight. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.MJD array([ 52307.5]) >>> a = Ticktock('2009-01-01T00:00:00', 'ISO') >>> a.MJD array([ 54832.00000579]) See Also ======== getUTC, getUNX, getRDT, getJD, getISO, getCDF, getTAI, getDOY, geteDOY, getAPT """ if self.data.attrs['dtype'] == 'MJD': # This should be the case from the constructor self.MJD = self.data return self.MJD # 1958-01-01T12:00 is MJD 36204.5 (days since 1858-11-17T00:00) self.MJD = _days1958(self.TAI, leaps='rubber') + 36204.5 return self.MJD
# -----------------------------------------------
[docs] def getUNX(self): """ a.UNX or a.getUNX() convert dtype data into Unix Time (Posix Time) seconds since 1970-1-1 (not counting leap seconds) Returns ``data`` if it was provided in UNX; otherwise always recalculates from the current value of ``TAI``, which will be created if necessary. Updates the ``UNX`` attribute. Returns ======= out : numpy array elapsed secs since 1970-1-1 (not counting leap secs) Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.UNX array([ 1.01265120e+09]) See Also ======== getUTC, getISO, getRDT, getJD, getMJD, getCDF, getTAI, getDOY, geteDOY, getAPT """ if self.data.attrs['dtype'] == 'UNX': # This should be the case from the constructor self.UNX = self.data return self.UNX naive_tai = _tai_real_to_naive(self.TAI) UNXofTAI0 = -378691200. unx = naive_tai + UNXofTAI0 self.UNX = spacepy.datamodel.dmarray(unx, attrs={'dtype': 'UNX'}) return self.UNX
# -----------------------------------------------
[docs] def getRDT(self): """ a.RDT or a.RDT() convert dtype data into Rata Die (lat.) Time, or elapsed days counting 0001-01-01 as day 1. This is a naive conversion: it ignores the existence of leapseconds for fractional days and ignores the conversion from Julian to Gregorian calendar, i.e. it assumes Gregorian calendar infinitely into the past. Returns ``data`` if it was provided in RDT; otherwise always recalculates from the current value of ``TAI``, which will be created if necessary. Updates the ``RDT`` attribute. Returns ======= out : numpy array elapsed days counting 1/1/1 as day 1. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.RDT array([ 730883.5]) See Also ======== getUTC, getUNX, getISO, getJD, getMJD, getCDF, getTAI, getDOY, geteDOY, getAPT """ if self.data.attrs['dtype'] == 'RDT': # This should be the case from the constructor self.RDT = self.data return self.RDT # RDT date at 1958-1-1T00 RDTTAI0 = 714780.0 RDT = _days1958(self.TAI, leaps='drop', midnight=True) + RDTTAI0 # RDT can represent 1582-10-5 through 1582-10-14, which do not exist. # So everything before 1582-10-15 (RDT day 577736) is ten days earlier. RDT[RDT < 577736.0] -= 10 self.RDT = spacepy.datamodel.dmarray(RDT, attrs={'dtype': 'RDT'}) return self.RDT
# -----------------------------------------------
[docs] def getUTC(self): """ a.UTC or a.getUTC() convert dtype data into UTC object a la datetime() Return value comes from (in priority order): 1. If ``data`` was provided in UTC, returns ``data``. 2. Else recalculates directly from ``data`` if it was provided in ISO. 3. Else calculates from current value of ``TAI``, which will be created if necessary. (``data`` is TAI, GPS, JD, MJD, RDT, CDF, UNX)). Updates the ``UTC`` attribute. Returns ======= out : list of datetime objects datetime object in UTC time Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.UTC [datetime.datetime(2002, 2, 2, 12, 0)] See Also ======== getISO, getUNX, getRDT, getJD, getMJD, getCDF, getTAI, getDOY, geteDOY, getAPT """ nTAI = len(self.data) # if already UTC, we are done, no conversion if self.data.attrs['dtype'].upper() == 'UTC': UTC = self.data elif self.data.attrs['dtype'].upper() == 'ISO': self.ISO = self.data _, UTC, _ = dtstr2iso(self.data, fmt=self._isofmt) elif self.data.attrs['dtype'].upper() in ( 'TAI', 'GPS', 'JD', 'MJD', 'RDT', 'CDF', 'UNX', 'APT'): TAI0 = datetime.datetime(1958, 1, 1, 0, 0, 0, 0) UTC = [datetime.timedelta( # Before 1582-10-15, UTC 10 days earlier than naive conversion # since those dates are Julian not Gregorian. seconds=float(tait - (864000 if tait < -11840601600.0 else 0))) + TAI0 for tait in self.TAI] for i in np.arange(nTAI): # This is the index of number of seconds to subtract from # "naive" UTC, not just TAI - UTC. # TAI of leap second does not have a new TAI - UTC, this # is because UTC seconds = 60, but need to subtract off # one more to make the UTC seconds = 59 in that case, thus # "flip" to next leap second count 1s earlier. idx = np.searchsorted(TAIleaps, self.TAI[i], side='right') - 1 UTC[i] = UTC[i] - datetime.timedelta(seconds=secs[idx] if idx > 0 else 0) if int(self.TAI[i]) == TAIleaps[idx]: # TAI is in leap second UTC[i] = UTC[i].replace( second=59, microsecond=999999) else: warnstr1 = 'Input data type {0} does not support calculation of UTC times'.format(self.data.attrs['dtype']) warnstr2 = 'Valid input dtypes are: {0}'.format( ', '.join([kk for kk in Ticktock._keylist if kk not in ['DOY', 'eDOY', 'leaps']])) raise TypeError('{0}\n{1}'.format(warnstr1, warnstr2)) self.UTC = spacepy.datamodel.dmarray(UTC, attrs={'dtype': 'UTC'}) return self.UTC
# -----------------------------------------------
[docs] def getGPS(self): """ a.GPS or a.getGPS() Return seconds since the GPS epoch (1980-1-6T00:00 UT) Always recalculates from the current value of ``TAI``, which will be created if necessary. Updates the ``GPS`` attribute. Returns ======= out : numpy array elapsed secs since 1980-1-6. Leap seconds are counted; i.e. there are no discontinuities. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.GPS dmarray([6.96686413e+08]) See Also ======== getUTC, getUNX, getRDT, getJD, getMJD, getCDF, getISO, getDOY, geteDOY, getAPT """ if self.data.attrs['dtype'] == 'GPS': # This should be the case from the constructor self.GPS = self.data return self.GPS GPS0 = 694656019 self.GPS = spacepy.datamodel.dmarray( self.TAI - GPS0, attrs={'dtype': 'GPS'}) return self.GPS
# -----------------------------------------------
[docs] def getAPT(self): """ a.APT or a.getAPT() Return AstroPy time object. Always recalculates from the current value of ``TAI``, which will be created if necessary. Updates the ``APT`` attribute. Returns ======= out : astropy.time.Time AstroPy Time object Notes ===== .. versionadded:: 0.2.2 Requires AstroPy 1.0. The returned value will be on the ``tai`` scale in ``gps`` format (unless the :class:`Ticktock` was created from a :class:`~astropy.time.Time` object, in which case it will be the original input.) See the :mod:`astropy.time` docs for conversion to other scales and formats. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.APT <Time object: scale='tai' format='gps' value=696686413.0> See Also ======== getUTC, getUNX, getRDT, getJD, getMJD, getCDF, getISO, getDOY, geteDOY, getGPS """ if self.data.attrs['dtype'] == 'APT': # This should be the case from the constructor self.APT = self.data return self.APT if not HAVE_ASTROPY: raise RuntimeError('Import of astropy.time failed.') GPS0 = 694656019 self.APT = astropy.time.Time( self.TAI - GPS0, scale='tai', format='gps') self.APT.attrs = {'dtype': 'APT'} return self.APT
# -----------------------------------------------
[docs] def getTAI(self): """ a.TAI or a.getTAI() return TAI (International Atomic Time), elapsed secs since 1958-1-1 (leap seconds are counted.) Ticktock's handling of TAI and UTC conversions treats the UTC second as always equal in length to the SI second (and thus TAI), ignoring frequency changes and fractional leap seconds from 1958 through 1972, i.e. the UTC to TAI offset is always treated as an integer, truncated (not rounded) from the value at the most recent leap second (or fraction thereof). Return value comes from (in priority order): 1. If ``data`` was provided in TAI, returns ``data``. 2. Else recalculates directly from ``data`` if it was provided in APT, CDF, GPS, ISO, JD, MJD, RDT, or UNX. 3. Else calculates from current value of ``UTC``, which will be created if necessary. Updates the ``TAI`` attribute; will also create the ``UTC`` and ``ISO`` attributes from ``data`` if input is in ``ISO`` (but will not overwrite an existing ``ISO`` or ``UTC``). This is for efficiency, as computation from ISO requires calculating UTC and makes creating a formatted ISO string easy. Returns ======= out : numpy array TAI as seconds since 1958-1-1. Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.TAI array([1391342432]) See Also ======== getUTC, getUNX, getRDT, getJD, getMJD, getCDF, getISO, getDOY, geteDOY, getAPT """ if self.data.attrs['dtype'] == 'TAI': # This should be the case from the constructor self.TAI = self.data return self.TAI if self.data.attrs['dtype'] == 'GPS': GPS0 = 694656019 self.TAI = spacepy.datamodel.dmarray( self.data + GPS0, attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'MJD': MJDofTAI0 = 36204.5 # Days-since-1958 is relative to noon self.TAI = spacepy.datamodel.dmarray( _days1958totai( np.require(self.data, dtype=np.float64) - MJDofTAI0, leaps='rubber', midnight=False), attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'JD': JDofTAI0 = 2436205.0 # Days-since-1958 is relative to noon self.TAI = spacepy.datamodel.dmarray( _days1958totai( np.require(self.data, dtype=np.float64) - JDofTAI0, leaps='rubber', midnight=False), attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'RDT': RDTofTAI0 = 714780. #RDT date of 1958-1-1T00 tai = _days1958totai( np.require(self.data, dtype=np.float64) - RDTofTAI0, leaps='drop', midnight=True) # Anything before 1582-10-5 has TAI ten days later than the # naive conversion, because RDT has ten days that are not in TAI. tai[tai < -11840601600.0] += 864000 self.TAI = spacepy.datamodel.dmarray(tai, attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'CDF': CDFofTAI0 = 61788528000000.0 # Naive TAI conversion tai = (self.data - CDFofTAI0) / 1.e3 tai = _tai_naive_to_real(tai) self.TAI = spacepy.datamodel.dmarray(tai, attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'UNX': UNXofTAI0 = -378691200. # Naive TAI tai = self.data - UNXofTAI0 tai = _tai_naive_to_real(tai) self.TAI = spacepy.datamodel.dmarray(tai, attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'APT': # AstroPy time doesn't support TAI directly, have to go relative # to GPS. But since it's a simple offset, just do the math here # instead of via the GPS attribute. GPS0 = 694656019 self.TAI = spacepy.datamodel.dmarray( self.data.gps + GPS0, attrs={'dtype': 'TAI'}) return self.TAI if self.data.attrs['dtype'] == 'ISO': isoout, UTC, offset = dtstr2iso(self.data, self._isofmt) if 'UTC' not in dir(self): self.UTC = spacepy.datamodel.dmarray( UTC, attrs={'dtype': 'UTC'}) if 'ISO' not in dir(self): self.ISO = spacepy.datamodel.dmarray( isoout, attrs={'dtype': 'ISO'}) else: UTC = self.UTC offset = None TAI0 = datetime.datetime(1958, 1, 1, 0, 0, 0, 0) leapsec = self.getleapsecs() TAItup = [utc - TAI0 + datetime.timedelta(seconds=int(ls)) for utc, ls in zip(UTC, leapsec)] if offset is not None: TAI = [tai.days * 86400 + tai.seconds + (tai.microseconds + offset[i])/ 1.e6 for i, tai in enumerate(TAItup)] else: TAI = [tai.days * 86400 + tai.seconds + tai.microseconds / 1.e6 for tai in TAItup] TAI = spacepy.datamodel.dmarray(TAI, attrs={'dtype': 'TAI'}) # 1582-10-5 through 1582-10-14 do not exist, so anything # before 1582-10-15 is 10 TAI days later than the naive conversion. TAI[TAI < -11840601600.0] += (86400 * 10) self.TAI = TAI return self.TAI
# -----------------------------------------------
[docs] def getISO(self): """ a.ISO or a.getISO() convert dtype data into ISO string Always recalculates from the current value of ``UTC``, which will be created if necessary. Applies leapsecond correction based on ``TAI``, also created as necessary. Updates the ``ISO`` attribute. Returns ======= out : list of strings date in ISO format Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.ISO dmarray(['2002-02-02T12:00:00']) See Also ======== getUTC, getUNX, getRDT, getJD, getMJD, getCDF, getTAI, getDOY, geteDOY, getAPT """ if self.data.attrs['dtype'] == 'ISO': # Convert the string directly. self.ISO = spacepy.datamodel.dmarray( dtstr2iso(self.data, fmt=self._isofmt)[0], attrs={'dtype': 'ISO'}) return self.ISO nTAI = len(self.data) self.TAI = self.getTAI() try: iso = [utc.strftime(self._isofmt) for utc in self.UTC] except ValueError: # Python before 3.3 fails on strftime before 1900. iso = [utc.replace(year=1900).strftime( self._isofmt.replace('%Y', str(utc.year))) for utc in self.UTC] self.ISO = spacepy.datamodel.dmarray(iso, attrs={'dtype': 'ISO'}) for i in range(nTAI): if int(self.TAI[i]) in TAIleaps: # UTC is 23:59:59.9999, get correct number of microseconds tmpdt = self.UTC[i].replace( microsecond=int((self.TAI[i] % 1) * 1e6)) # And fudge the second a, b, c = tmpdt.strftime(self._isofmt).split(':') cnew = c.replace('59', '60') self.ISO[i] = a + ':' + b + ':' + cnew return self.ISO
# -----------------------------------------------
[docs] def getleapsecs(self): """ a.leaps or a.getleapsecs() retrieve leapseconds from lookup table, used in getTAI Always recalculates from current value of ``TAI`` if ``data`` is dtype ``TAI``, otherwise from the current value of ``UTC``, which will be created if necessary. Updates the ``leaps`` attribute. Returns ======= out : numpy array leap seconds Examples ======== >>> a = Ticktock('2002-02-02T12:00:00', 'ISO') >>> a.leaps array([32]) See Also ======== getTAI """ if self.data.attrs['dtype'] == 'TAI': # TAIleaps contains the TAI which IS the leap second. # The leap second count increments in the NEXT second. # (TAI - UTC changes at end of leap second.) # So find the latest index where the TAI-after-leap-second # is less than current TAI (i.e. we are not after leap second yet). idx = np.searchsorted(TAIleaps + 1, self.data, side='right') - 1 return secs[idx] tup = self.UTC # check if array: if isinstance(tup, datetime.datetime): # not an array of objects tup = [tup] nTAI = 1 aflag = False else: nTAI = len(tup) aflag = True # convert them into a time tuple and find the correct leap seconds self.TAIleaps = TAIleaps leaps = [secs[0]] * nTAI leap_dates = [datetime.datetime(int(y), int(m), int(d)) for y, m, d, s in zip(year, mon, day, secs)] for i, itup in enumerate(tup): ind = bisect.bisect_right(leap_dates, tup[i]) leaps[i] = secs[ind - 1] if ind > 0 else 0 ## ldatetime = datetime.datetime # avoid an expensive lookup below ## for i, itup in enumerate(tup): ## for y,m,d,s in zip(year, mon, day, secs): ## if tup[i] >= ldatetime(int(y),int(m),int(d)): ## leaps[i] = s ## else: ## break # if datetime.datetime(1971,12,31) > tup[0]: # print "WARNING: date before 1972/1/1; leap seconds are by fractions off" if aflag == False: self.leaps = int(leaps[0]) return int(leaps[0]) # if you want to allow fractional leap seconds, remove 'int' here else: self.leaps = np.array(leaps, dtype=int) return self.leaps
# -----------------------------------------------
[docs] @classmethod def now(cls): """ Create Ticktock with the current UTC time. Equivalent to datetime.utcnow() .. versionchanged:: 0.2.2 This now returns a UTC time; previously it returned a Ticktock UTC object, but in the local timezone, which made all conversions incorrect. Returns ======= out : ticktock Ticktock object with the current time, equivalent to datetime.utcnow() See Also ======== datetime.datetime.now, datetime.datetime.utcnow """ warnings.warn('now() returns UTC time as of 0.2.2.', DeprecationWarning) dt = datetime.datetime.utcnow() return Ticktock(dt, 'UTC')
# -----------------------------------------------
[docs] @classmethod def today(cls): """ Create Ticktock with the current UTC date and time set to 00:00:00 Similar to date.today() with time included but in UTC and with the time included (zero hours, minutes, seconds) .. versionchanged:: 0.2.2 This now returns the UTC day; previously it returned a Ticktock UTC object, but in the local timezone, which made all conversions incorrect. Returns ======= out : ticktock Ticktock object with the current UTC day See Also ======== datetime.date.today """ warnings.warn('today() returns UTC day as of 0.2.2.', DeprecationWarning) dt = datetime.datetime.utcnow() dt = dt.replace(hour=0, minute=0, second=0, microsecond=0) return Ticktock(dt, 'UTC')
# ----------------------------------------------- # End of Ticktock class # -----------------------------------------------
[docs]def doy2date(year, doy, dtobj=False, flAns=False): """ convert integer day-of-year doy into a month and day after http://pleac.sourceforge.net/pleac_python/datesandtimes.html Parameters ========== year : int or array of int year doy : int or array of int day of year Returns ======= month : int or array of int month as integer number day : int or array of int as integer number Examples ======== >>> month, day = doy2date(2002, 186) >>> dts = doy2date([2002]*4, range(186,190), dtobj=True) See Also ======== Ticktock.getDOY """ try: n_year = len(year) except TypeError: n_year = -1 # Special case: this is a scalar try: n_doy = len(doy) except TypeError: n_doy = -1 if n_doy != n_year: raise ValueError('Day of year and year must have same length') if n_year == -1: if doy < 1: raise ValueError( 'Day-of-Year less than 1 detected: DOY starts from 1') if not flAns: dt = datetime.datetime(int(year), 1, 1) + \ datetime.timedelta(days=int(doy) - 1) else: dt = datetime.datetime(year, 1, 1) + \ datetime.timedelta(days=float(doy) - 1) if dtobj: return dt else: return dt.month, dt.day if min(doy) < 1: raise ValueError('Day-of-Year less than 1 detected: DOY starts from 1') if flAns: dateobj = spacepy.datamodel.dmarray([datetime.datetime(year[i], 1, 1) + datetime.timedelta(days=float(doy[i]) - 1) for i in range(n_year)]) else: dateobj = spacepy.datamodel.dmarray([datetime.datetime(int(year[i]), 1, 1) + datetime.timedelta(days=int(doy[i]) - 1) for i in range(n_year)]) if dtobj: return dateobj else: return (spacepy.datamodel.dmarray([dt.month for dt in dateobj]), spacepy.datamodel.dmarray([dt.day for dt in dateobj]))
# -----------------------------------------------
[docs]def tickrange(start, end, deltadays, dtype=None): """ return a Ticktock range given the start, end, and delta Parameters ========== start : string or number start time (ISO standard string and UTC/datetime do not require a dtype) end : string or number last possible time in series (excluded unless end=start+n*step for integer n) deltadays : float or timedelta step in units of days (float); or datetime timedelta object dtype : string (optional) data type for start, end; e.g. ISO, UTC, RTD, etc. see Ticktock for all options Returns ======= out : Ticktock instance ticks Examples ======== >>> ticks = st.tickrange('2002-02-01T00:00:00', '2002-02-10T00:00:00', deltadays = 1) >>> ticks Ticktock( ['2002-02-01T00:00:00', '2002-02-02T00:00:00', '2002-02-03T00:00:00', '2002-02-04T00:00:00'] , dtype=ISO) See Also ======== Ticktock """ Tstart = Ticktock(start, dtype) Tend = Ticktock(end, dtype) diff = Tend.UTC[0] - Tstart.UTC[0] dmusec, dsec = diff.microseconds / 86400000000., diff.seconds / 86400. if isinstance(deltadays, datetime.timedelta): musec, sec = deltadays.microseconds / 86400000000., deltadays.seconds / 86400. deltat = musec + sec + deltadays.days nticks = int((dmusec + dsec + diff.days) / deltat + 1) trange = [Tstart.UTC[0] + deltadays * n for n in range(nticks)] else: nticks = int((dmusec + dsec + diff.days) / float(deltadays) + 1) trange = [Tstart.UTC[0] + datetime.timedelta(days=deltadays) * n for n in range(nticks)] ticks = Ticktock(trange, 'UTC') return ticks
[docs]def dtstr2iso(dtstr, fmt='%Y-%m-%dT%H:%M:%S'): """Convert a datetime string to a standard format Attempts to maintain leap second representation while converting time strings to the specified format (by default, ISO8601-like.) Only handles a single positive leap second; negative leap seconds require no special handling and policy is for UTC-UT1 not to exceed 0.9. Parameters ========== dtstr : sequence of str Date + time representation, format is fairly open. Returns ======= isostr : array of str Representation of `dtstr` formatted according to `fmt`. Always a new sequence even if contents are identical to `dtstr`. UTC : array of datetime.datetime The closest-possible rendering of UTC time before or equal to `dtstr`. offset : array of int Amount (in microseconds) to add to `UTC` to get the real time. Other Parameters ================ fmt : str, optional Format appropriate for :meth:`~datetime.datetime.strftime` for rendering the output time. """ indtstr = dtstr # Will be editing this, so force it to own its data (but keep copy!) dtstr = np.require(indtstr, dtype='S' if str is bytes else 'U', requirements='O') dtstr = dtstr.copy() if dtstr is indtstr else dtstr # Replace leapsecond with a valid "59" # Indices of every place that might be leap second # Leap second is sec==60, must come at LEAST after YYMMDDHHMM, if # being very terse, but this still will avoid YYYY-060 lsidx = np.char.rfind(dtstr, '60') >= 10 if lsidx.shape == (): lsidx = lsidx.reshape((1,)) #atleast1d is only in new numpy leapidx = np.transpose(np.nonzero(lsidx)) # Add offset to datetime value to get actual UTC, # in integer microseconds. offset = np.zeros(shape=dtstr.shape, dtype=np.uint32) # Actual leap second indices (not just suspected) realleap = [] for j, idx in enumerate(leapidx): # Should be short list, so loop it # Get the index to scalar if necessary i = tuple(idx) if dtstr.shape else () # The leap second must be preceded by at least 10 char (above), # followed by either nothing or fractional seconds, # preceded by 59 minutes (and optional separator) # It must also not be preceded by a . (to avoid replacing # 60 milliseconds with 59 milliseconds). dtstr[i], count = re.subn( r'^([^\.]{8,}59[^\d]?)60((?:\.\d+)?)$', r'\g<1>59\g<2>', dtstr[i]) # Doing this subtracted one second. if count: realleap.append(idx) offset[i] = 1e6 # Cut index of leap seconds down to real ones. leapidx = np.array(realleap) # try a few special cases that are faster than dateutil.parser strfmts = ['%Y-%m-%dT%H:%M:%S', '%Y-%m-%dT%H:%M:%SZ', '%Y-%m-%d', '%Y%m%d', '%Y%m%d %H:%M:%S'] if fmt not in strfmts: strfmts.insert(0, fmt) for strfmt in strfmts: try: UTC = np.frompyfunc( lambda x: datetime.datetime.strptime(x, strfmt), 1, 1)(dtstr) break except ValueError: continue else: UTC = np.frompyfunc(dup.parse, 1, 1)(dtstr) otypes = ['S' if str is bytes else 'U'] try: isostr = np.vectorize(lambda x: x.strftime(fmt), otypes=otypes)(UTC) except ValueError: # Python before 3.3 fails on strftime before 1900. isostr = np.vectorize( lambda x: x.replace(year=1900).strftime(fmt.replace( '%Y', str(x.year))), otypes=otypes)(UTC) # Check that leap seconds are actually valid if len(leapidx): # Day that ends in leap second *entry* (may not be leap second) # Unbelievably these are read as floats, and other places rely on that. leapsecday = np.array([ datetime.date(int(y), int(m), int(d))- datetime.timedelta(days=1) for y, m , d in zip(year, mon, day)]) # Find only those leap seconds that are really changes idx = np.nonzero(np.diff(np.concatenate(([0], secs))))[0] leapsecday = leapsecday[idx] if dtstr.shape == (): if UTC.date() not in leapsecday: raise ValueError('{} is not a valid leapsecond.'.format( indtstr)) else: utcday = np.frompyfunc(lambda x: x.date(), 1, 1)(UTC[leapidx]) # Last day w/leapsecond that comes before supposed leapsec day closestday = np.clip( np.searchsorted(leapsecday, utcday), None, len(leapsecday) - 1) bad = utcday != leapsecday[closestday] if bad.any(): raise ValueError('{} is not a valid leapsecond.'.format( indtstr[leapidx[bad][0]])) #Peg all leap seconds to the end of the previous second for idx in leapidx: i = tuple(idx) if dtstr.shape else () if not offset[i]: continue # The time string is off by one second. if i == (): # Scalar input. isostr = UTC.strftime(fmt.replace('%S', '60')) us = UTC.microsecond UTC = UTC + datetime.timedelta(microseconds=(1e6 - us - 1)) offset = us + 1 else: isostr[i] = UTC[i].strftime(fmt.replace('%S', '60')) us = UTC[i].microsecond UTC[i] = UTC[i] + datetime.timedelta(microseconds=(1e6 - us - 1)) offset[i] = us + 1 return isostr, UTC, offset
[docs]def sec2hms(sec, rounding=True, days=False, dtobj=False): """Convert seconds of day to hours, minutes, seconds Parameters ========== sec : float Seconds of day Other Parameters ================ rounding : boolean set for integer seconds days : boolean set to wrap around day (i.e. modulo 86400) dtobj : boolean set to return a timedelta object Returns ======= out : [hours, minutes, seconds] or datetime.timedelta """ if rounding: sec = int(round(sec)) if not days: if sec > 86400: warnings.warn("Number of seconds > seconds in day. " "Try days keyword.") else: sec %= 86400 if dtobj: # no need to do the computation return datetime.timedelta(seconds=sec) else: hours = sec // 3600 minutes = ((sec - hours * 3600) // 60) % 60 seconds = sec % 60 return [hours, minutes, seconds]
def no_tzinfo(dt): """ take in an arraylike of datetime objects and return them without any tzinfo Parameters ========== dt : iterable iterable of datetime.datetime objects Returns ======= out : list list of datetime.datetime without tzinfo """ returnclass = type(dt) try: retval = [val.replace(tzinfo=None) for val in dt] except TypeError: # was not an iterable, assume datetime return dt.replace(tzinfo=None) #special case: numpy ndarray - dmarray and masked array work, but not ndarray if returnclass is not np.ndarray: retval = returnclass(retval) if hasattr(dt, 'attrs') and hasattr(retval, 'attrs'): retval.attrs.update(dt.attrs) return retval else: return np.asarray(retval)
[docs]def leapyear(year, numdays=False): """ return an array of boolean leap year, a lot faster than the mod method that is normally seen Parameters ========== year : array_like array of years numdays : boolean (optional) optionally return the number of days in the year Returns ======= out : numpy array an array of boolean leap year, or array of number of days Examples ======== >>> import numpy >>> import spacepy.time >>> spacepy.time.leapyear(numpy.arange(15)+1998) [False, False, True, False, False, False, True, False, False, False, True, False, False, False, True] """ year = np.asanyarray(year) mask400 = (year % 400) == 0 mask100 = (year % 100) == 0 mask4 = (year % 4) == 0 isleap = (mask400 | mask4) & (~mask100 | mask400) if numdays: isleap = isleap.astype(int) + 365 return isleap
[docs]def randomDate(dt1, dt2, N=1, tzinfo=False, sorted=False): """ Return a (or many) random datetimes between two given dates Convention used is dt1 <= rand < dt2. Leap second times will not be returned. Parameters ========== dt1 : datetime.datetime start date for the the random date dt2 : datetime.datetime stop date for the the random date Other Parameters ================ N : int (optional) the number of random dates to generate (defualt=1) tzinfo : bool (optional) maintain the tzinfo of the input datetimes (default=False) sorted : bool (optional) return the times sorted (default=False) Returns ======= out : datetime.datetime or numpy.ndarray of datetime.datetime the new time for the next call to EventTimer Examples ======== """ if dt1.tzinfo != dt2.tzinfo: raise ValueError('tzinfo for the input and output datetimes must match') tt = Ticktock([dt1, dt2]).RDT rnd_tn = np.random.uniform(tt[0], tt[1], size=N) rnd_t = Ticktock(rnd_tn, dtype='RDT') if sorted: rnd_t.sort() rnd_t = np.asarray([val.replace(tzinfo=dt1.tzinfo if tzinfo else None) for val in rnd_t.UTC]) return rnd_t
def extract_YYYYMMDD(filename): """ go through a string and extract the first valid YYYYMMDD as a datetime Parameters ========== filename : str string to parse for a YYYYMMDD format Returns ======= out : (None, datetime.datetime) the datetime found in the string or None """ # return a datetime if there is one from YYYYMMDD # be picky so don't match random numbers (1950-2049) m = re.search(r"(19[5-9]|20[0-4])\d(0\d|1[0-2])([0-2]\d|3[01])", os.path.basename(filename)) if not m: return None else: return datetime.datetime.strptime(m.group(), '%Y%m%d') def valid_YYYYMMDD(inval): """ if inval is valid YYYYMMDD return True, False otherwise """ if re.search(r"(19[5-9]|20[0-4])\d(0\d|1[0-2])([0-2]\d|3[01])", inval): return True else: return False def _leapsgood(now, filetime, lastleap): """Determines if leap seconds are up-to-date Parameters ---------- now : datetime.datetime Current time (UTC assumed) filetime : datetime.datetime Timestamp (mtime) of the leapsecond file lastleap : datetime.datetime Last leapsecond in the file (really the moment after the leap) Returns ------- bool True if leap second file is up-to-date, False if might not be. """ # There cannot be a leap second until AFTER the ls after the # bulletin (e.g. once past 1 Jan, know the next possible is next 1 Jan) goodto_mtime = datetime.datetime( filetime.year + 1, 7 if filetime.month > 6 else 1, 1) # The next possible leapsecond is 6mo after the previous known leap, # which is (technically just before) 1 Jan or 1 July goodto_ls = datetime.datetime( lastleap.year + int(lastleap.month > 6), 1 if lastleap.month > 6 else 7, 1) goodto = max(goodto_ls, goodto_mtime) return now < goodto def _read_leaps(oldstyle=False): """Read leapseconds in from spacepy tai-utc file Populates module-global variables with leapsecond information: secs, year, mon, day, TAIleaps. Called on import of this module. Other Parameters ---------------- oldstyle : bool .. versionadded:: 0.2.3 Treat leapseconds as in SpacePy 0.2.2 (default False). Default is to ignore the file contents through 1 Jan 1972 and use SpacePy's own list of integral leapseconds, which uses the post-1972 standard (add a leapsecond on January 1/July 1 if UTC - UT1 > 0.4s). """ global secs, year, mon, day, TAIleaps # load current file fname = os.path.join(spacepy.DOT_FLN, 'data', 'tai-utc.dat') try: with open(fname) as fh: text = fh.readlines() mtime = datetime.datetime(*time.gmtime(os.path.getmtime(fname))[:6]) except IOError: warnings.warn('Cannot read leapsecond file. Use' ' spacepy.toolbox.update(leapsecs=True).') text = [] # Use built-in pre-1972 leaps mtime = None # Some files have a "last checked" line at the top if text and text[0].startswith('Checked'): del text[0] months = np.array(['JAN', 'FEB', 'MAR', 'APR', 'MAY', 'JUN', 'JUL', 'AUG', 'SEP', 'OCT', 'NOV', 'DEC']) if not oldstyle: # Use internal integral leapsecond count until 1972 keep_idx = next((i for i, l in enumerate(text) if int(l[:5]) > 1972 or int(l[:5]) == 1972 and l[6:9] == 'JUL'), len(text)) leaps = [(1959, 1, 36569), (1961, 1, 37300), (1963, 7, 38211), (1965, 1, 38761), (1966, 7, 39307), (1967, 7, 39672), (1968, 7, 40038), (1969, 7, 40403), (1970, 7, 40768), (1971, 7, 41133)] # 1972 Jan NOT a leapsecond in this formulation; TAI-UTC=10s at 71 Jul text = [ ' {Y} {M} 1 =JD {JD} TAI-UTC={L:12.7f} S ' '+ (MJD - 00000.) X 0.0000000 S ACTUAL\n'.format( Y=l[0], M=months[l[1] - 1], JD=l[2] + 2400000.5, L=i + 1) for i, l in enumerate(leaps)] \ + text[keep_idx:] secs = np.zeros(len(text)) year = np.zeros(len(text)) mon = np.zeros(len(text)) day = np.zeros(len(text)) for line, i in zip(text, np.arange(len(secs))): # Round float seconds (0.5 always rounds up.) secs[i] = int(float(line.split()[6]) + 0.5) year[i] = int(line.split()[0]) mon[i] = int(np.where(months == line.split()[1])[0][0] + 1) day[i] = int(line.split()[2]) # Check for out of date. The leap second bulletin comes every # six months, and that contains information through the following # leap second (end of June/Dec) if mtime is not None and spacepy.config['enable_old_data_warning'] \ and not _leapsgood( datetime.datetime.utcnow(), mtime, datetime.datetime(int(year[-1]), int(mon[-1]), int(day[-1]))): warnings.warn('Leapseconds may be out of date.' ' Use spacepy.toolbox.update(leapsecs=True)') TAIleaps = np.zeros(len(secs)) TAItup = [''] * len(secs) TAI0 = datetime.datetime(1958, 1, 1, 0, 0, 0, 0) for i in np.arange(len(secs)): TAItup[i] = datetime.datetime(int(year[i]), int(mon[i]), int(day[i])) - TAI0 + datetime.timedelta( seconds=int(secs[i]) - 1) TAIleaps[i] = TAItup[i].days * 86400 + TAItup[i].seconds + TAItup[i].microseconds / 1.e6 def _days1958(tai, leaps='rubber', midnight=False): """Calculate days and fractional days since 1958-01-01T12:00 This is basically a Julian Date but baselined from the start of TAI. Since it is calculated from TAI, using this as day 0 maximizes the resolution of the resulting values. Parameters ========== tai : sequence of float TAI seconds (i.e. continuous SI seconds relative to 1958-01-01T00:00) leaps : str, optional How to treat days with leapseconds. Since the Julian Date runs noon (inclusive) to noon (exclusive), this affects the back half of the date with the leapsecond, and the first half of the next. Accepted values are: rubber Consider these days to be 86401 seconds long and thus treat each second as a slightly smaller fraction of the day, so that all times are represented and evently spaced within the day. I.e. the day is "stretched" across more seconds, by analogy with the "rubber second" of 1960-1972. (default) drop Treate time as a flow of SI seconds that is suspended during leapseconds, i.e. leapsecond time does not exist. Time during leap seconds is pinned to the last microsecond of the previous second. continuous Treat time as a continuous flow of SI seconds with days always 86400 seconds long. This is the proper handling for true Julian Dates, i.e. JD(TAI), as recommended by IAU General Assembly XXIII, resolution B1. Returns ======= sequence of float Days, including fraction, relative to 1958-01-01T12:00 Other parameters ================ midnight : bool, optional Start the day at midnight instead of noon. This affects the allocation of leap seconds, and of course days are relative to 1958-01-01T00:00 """ off = 0. if midnight else 43200. # Offset from midnight # Shift to time-since-noon, if desired (also makes copy), call delta-TAI dtai = np.require(tai, dtype=np.float64) - off leap_dtai, taiutc = _changed_leaps() leap_dtai -= off # delta-TAI of start of leap second if leaps in ('rubber', 'drop'): # Index of leapsecond equal to or before each time record lidx = np.searchsorted(leap_dtai, dtai, side='right') - 1 elif leaps != 'continuous': raise ValueError('leaps handling {} not recognized.'.format(leaps)) if leaps == 'rubber': # d-TAI of start-of-day (noon or midnight) before/after each leapsecond # because leapseconds always start at 23:59:60 (43200s after noon). if midnight: daystart_before_leap = leap_dtai - 86400 daystart_after_leap = leap_dtai + 1 else: daystart_before_leap = leap_dtai - 43200 daystart_after_leap = leap_dtai + 43201 # Closest leapsecond-day-start before record. # May be greater than lidx, if near but before leapsecond ldidx = np.searchsorted(daystart_before_leap, dtai, side='right') - 1 # All records that happen on leap second days. leap_sec_day = (daystart_before_leap[ldidx] <= dtai) \ & (dtai < daystart_after_leap[ldidx]) # Save SSD on leap second days, about to be destroyed if leap_sec_day.shape == (): # Scalar ssd_leap_sec_day = dtai - daystart_before_leap[ldidx] else: ssd_leap_sec_day = dtai[leap_sec_day] \ - daystart_before_leap[ldidx[leap_sec_day]] # Destroy leap seconds, so days always break at 86400s dtai -= taiutc[lidx] elif leaps == 'drop': # Where in a leapsecond, pin to previous second inleap = dtai - leap_dtai[lidx] < 1 if inleap.shape == (): # Scalar if inleap: dtai = np.floor(dtai) - .000001 lidx -= 1 # Associated with previous TAI - UTC else: dtai[inleap] = np.floor(dtai[inleap]) - .000001 # Already subtracted ~1sec, now associated with previous TAI - UTC lidx[inleap] -= 1 # Remove all of the TAI seconds that "disappeared". dtai -= taiutc[lidx] day = np.floor(dtai / 86400) ssd = dtai - day * 86400 # Mod does wrong thing if negative. if leaps == 'rubber': # Patch back in the SSD on leap-second days. if leap_sec_day.shape == (): # Scalar if leap_sec_day: ssd = ssd_leap_sec_day else: ssd[leap_sec_day] = ssd_leap_sec_day daylen = np.choose(leap_sec_day, (86400., 86401.)) else: daylen = 86400 return day + ssd / daylen def _days1958totai(days, leaps='rubber', midnight=False): """Calculate TAI from days and fractional days since 1958-01-01T12:00 Input is basically a Julian Date but baselined from the start of TAI. Since it is calculated from TAI, using this as day 0 maximizes the resolution of the resulting values. Parameters ========== days : sequence of float Days, including fraction, relative to 1958-01-01T12:00 leaps : str, optional How to treat days with leapseconds. Since the Julian Date runs noon (inclusive) to noon (exclusive), this affects the back half of the date with the leapsecond, and the first half of the next. Accepted values are: rubber Consider these days to be 86401 seconds long and thus treat each second as a slightly smaller fraction of the day, so that all times are represented and evently spaced within the day. I.e. the day is "stretched" across more seconds, by analogy with the "rubber second" of 1960-1972. (default) drop Treate time as a flow of SI seconds that is suspended during leapseconds, i.e. leapsecond time does not exist. Time during leap seconds is not recovered (no output time during leaps.) continuous Treat time as a continuous flow of SI seconds with days always 86400 seconds long. This is the proper handling for true Julian Dates, i.e. JD(TAI), as recommended by IAU General Assembly XXIII, resolution B1. Returns ======= sequence of float TAI seconds (i.e. continuous SI seconds relative to 1958-01-01T00:00) Other parameters ================ midnight : bool, optional Start the day at midnight instead of noon. This affects the allocation of leap seconds, and of course `days` are relative to 1958-01-01T00:00 """ days = np.require(days, dtype=np.float64) off = 0. if midnight else 43200. # Offset from midnight if leaps == 'continuous': # Very simple case. return days * 86400 + off elif leaps not in ('rubber', 'drop'): raise ValueError('leaps handling {} not recognized.'.format(leaps)) leap_tai, taiutc = _changed_leaps() leap_dtai = leap_tai - off # delta-TAI of start of leap second # Days with leap second. Leap second is always late in day: end # of day if doing midnight-to-midnight, halfway if noon-to-noon. leap_day = np.floor((leap_dtai - taiutc) / 86400) # Closest leapsecond-day before record. ldidx = np.searchsorted(leap_day, days) - 1 # All records that happen on leap second days. leap_sec_day = (ldidx > 0) \ & (leap_day[ldidx] <= days) \ & (days < leap_day[ldidx] + 1) if leaps == 'rubber': daylen = np.choose(leap_sec_day, (86400., 86401.)) else: daylen = 86400. # Naive TAI at start of day (no leapsecond corrections). dayint = np.floor(days) daystart = dayint * 86400 # Seconds from the start of naive day. ssd = (days - dayint) * daylen # Mod does wrong thing if negative # Index TAIUTC for every input. Start of day w/LS is still previous value. taiutcidx = ldidx - np.require(leap_sec_day, dtype=np.intp) # Keep small number adds together: ssd and leapseconds. taiout = daystart + (ssd + off + taiutc[taiutcidx]) if leaps == 'drop': # Any times after leap-second skip need an extra second. taiutcidx += np.require(leap_sec_day & (taiout >= leap_tai[ldidx]), dtype=np.intp) # Recalculate to keep small-number addition together taiout = daystart + (ssd + off + taiutc[taiutcidx]) return taiout def _changed_leaps(): """Find only those times where leap seconds actually changed There are leap second records where there is no actual change in leap seconds, and there isn't a record for TAI-UTC == 0; this adds a record (where the TAI of the leap second time is -inf) and eliminated those with no actual change. Returns ======= leap_tai : sequence of float TAI of the start of every leap second. taiutc : sequence of float TAI - UTC at the end of the leap second. """ # Find only those leap seconds that are really changes idx = np.nonzero(np.diff(np.concatenate(([0], secs))))[0] leap_tai = TAIleaps[idx] # TAI of start of leap second taiutc = secs[idx] # TAI - UTC after leap second (same index as leap_tai) # Add fake TAI - UTC = 0 at the Big Bang leap_tai = np.concatenate(([-np.inf], leap_tai)) taiutc = np.concatenate(([0], taiutc)) return leap_tai, taiutc def _tai_naive_to_real(tai): """Convert naive TAI to real TAI Convert a TAI on a continuous timescale that skips over leapseconds and is unaware of calendar conversion to actual TAI, which includes leapseconds and is continuous across the Gregorian conversion (naive has a gap, i.e. dates it can represent that don't exist.) Parameters ========== tai : sequence of float Naive TAI Returns ======= tai : sequence of float TAI """ # This is the ACTUAL TAI and TAI-UTC at the end of that TAI leap_tai, taiutc = _changed_leaps() # Naive TAI and index that corresponds to TAI - UTC naive_leap_tai = leap_tai - taiutc + 1 taiutcidx = np.searchsorted(naive_leap_tai, tai, side='right') - 1 realtai = tai + taiutc[taiutcidx] # Anything before 1582-10-5 has TAI ten days later than the # naive conversion, because naive has ten days that are not in TAI. realtai[realtai < -11840601600.0] += 864000 return realtai def _tai_real_to_naive(tai): """Convert naive TAI to actual TAI Convert actual TAI, which includes leapseconds and is continuous across the Gregorian calendar conversion, to naive TAI, which skips over leapseconds and has times in the calendar conversion that do not actually exist. Parameters ========== tai : sequence of float TAI Returns ======= tai : sequence of float Naive TAI """ # ACTUAL TAI and TAI-UTC at the end of that TAI leap_tai, taiutc = _changed_leaps() # Points to largest leap-TAI less-than input TAI, thus also TAI-UTC lidx = np.searchsorted(leap_tai, tai, side='right') - 1 # Records in a leap second inleap = tai < leap_tai[lidx] + np.diff(taiutc)[lidx - 1] naive_tai = np.choose(inleap, ( tai - taiutc[lidx], # Just subtract off LS np.floor(tai) + (.999 - taiutc[lidx]) # Peg to end of sec )) # Anything before 1582-10-15 needs to skip 10 CDF days backward, # since naive has ten days that are not in TAI naive_tai[tai < -11839737600.0] -= 864000 return naive_tai _read_leaps()