I have a data file containing timestamps like “1331856000000”. Unfortunately, I don’t have a lot of documentation for the format, so I’m not sure how the timestamp is formatted. I’ve tried Python’s standard datetime.fromordinal() and datetime.fromtimestamp() and a few others, but nothing matches. I’m pretty sure that particular number corresponds to the current date (e.g. 2012-3-16), but not much more.

How do I convert this number to a datetime?

datetime.datetime.fromtimestamp() is correct, except you are probably having timestamp in miliseconds (like in JavaScript), but fromtimestamp() expects Unix timestamp, in seconds.

Do it like that:

>>> import datetime
>>> your_timestamp = 1331856000000
>>> date = datetime.datetime.fromtimestamp(your_timestamp / 1e3)

and the result is:

>>> date
datetime.datetime(2012, 3, 16, 1, 0)

Does it answer your question?

EDIT: J.F. Sebastian correctly suggested to use true division by 1e3 (float 1000). The difference is significant, if you would like to get precise results, thus I changed my answer. The difference results from the default behaviour of Python 2.x, which always returns int when dividing (using / operator) int by int (this is called floor division). By replacing the divisor 1000 (being an int) with the 1e3 divisor (being representation of 1000 as float) or with float(1000) (or 1000. etc.), the division becomes true division. Python 2.x returns float when dividing int by float, float by int, float by float etc. And when there is some fractional part in the timestamp passed to fromtimestamp() method, this method’s result also contains information about that fractional part (as the number of microseconds).

I have dt = datetime(2013,9,1,11), and I would like to get a Unix timestamp of this datetime object.

When I do (dt - datetime(1970,1,1)).total_seconds() I got the timestamp 1378033200.

When converting it back using datetime.fromtimestamp I got datetime.datetime(2013, 9, 1, 6, 0).

The hour doesn’t match. What did I miss here?

What you missed here is timezones.

Presumably you’ve five hours off UTC, so 2013-09-01T11:00:00 local and 2013-09-01T06:00:00Z are the same time.

You need to read the top of the datetime docs, which explain about timezones and “naive” and “aware” objects.

If your original naive datetime was UTC, the way to recover it is to use utcfromtimestamp instead of fromtimestamp.

On the other hand, if your original naive datetime was local, you shouldn’t have subtracted a UTC timestamp from it in the first place; use datetime.fromtimestamp(0) instead.

Or, if you had an aware datetime object, you need to either use a local (aware) epoch on both sides, or explicitly convert to and from UTC.

If you have, or can upgrade to, Python 3.3 or later, you can avoid all of these problems by just using the timestamp method instead of trying to figure out how to do it yourself. And even if you don’t, you may want to consider borrowing its source code.

(And if you can wait for Python 3.4, it looks like PEP 341 is likely to make it into the final release, which means all of the stuff J.F. Sebastian and I were talking about in the comments should be doable with just the stdlib, and working the same way on both Unix and Windows.)

solution is

import time
import datetime
d = datetime.date(2015,1,5)

unixtime = time.mktime(d.timetuple())

If you want to convert a python datetime to seconds since epoch you should do it explicitly:

>>> import datetime
>>> datetime.datetime(2012,04,01,0,0).strftime('%s')
'1333234800'
>>> (datetime.datetime(2012,04,01,0,0) - datetime.datetime(1970,1,1)).total_seconds()
1333238400.0

In Python 3.3+ you can use timestamp() instead:

>>> import datetime
>>> datetime.datetime(2012,4,1,0,0).timestamp()
1333234800.0

Rather than this expression to create a POSIX timestamp from dt,

(dt - datetime(1970,1,1)).total_seconds()

Use this:

int(dt.strftime("%s"))

I get the right answer in your example using the second method.

EDIT: Some followup… After some comments (see below), I was curious about the lack of support or documentation for %s in strftime. Here’s what I found:

In the Python source for datetime and time, the string STRFTIME_FORMAT_CODES tells us:

"Other codes may be available on your platform.
 See documentation for the C library strftime function."

So now if we man strftime (on BSD systems such as Mac OS X), you’ll find support for %s:

"%s is replaced by the number of seconds since the Epoch, UTC (see mktime(3))."

Anyways, that’s why %s works on the systems it does. But there are better solutions to OP’s problem (that take timezones into account). See @abarnert’s accepted answer here.

For working with UTC timezones:

time_stamp = calendar.timegm(dt.timetuple())

datetime.utcfromtimestamp(time_stamp)

You’ve missed the time zone info (already answered, agreed)

arrow package allows to avoid this torture with datetimes; It is already written, tested, pypi-published, cross-python (2.6 — 3.xx).

All you need: pip install arrow (or add to dependencies)

Solution for your case

dt = datetime(2013,9,1,11)
arrow.get(dt).timestamp
# >>> 1378033200

bc = arrow.get(1378033200).datetime
print(bc)
# >>> datetime.datetime(2013, 9, 1, 11, 0, tzinfo=tzutc())
print(bc.isoformat())
# >>> '2013-09-01T11:00:00+00:00'

If your datetime object represents UTC time, don’t use time.mktime, as it assumes the tuple is in your local timezone. Instead, use calendar.timegm:

>>> import datetime, calendar
>>> d = datetime.datetime(1970, 1, 1, 0, 1, 0)
>>> calendar.timegm(d.timetuple())
60

Well, when converting TO unix timestamp, python is basically assuming UTC, but while converting back it will give you a date converted to your local timezone.

See this question/answer; Get timezone used by datetime.datetime.fromtimestamp()

def dt2ts(dt, utc=False):
    if utc:
        return calendar.timegm(dt.timetuple())
    if dt.tzinfo is None:
        return int(time.mktime(dt.timetuple()))
    utc_dt = dt.astimezone(tz.tzutc()).timetuple()
    return calendar.timegm(utc_dt)

If you want UTC timestamp :time.mktime just for local dt .Use calendar.timegm is safe but dt must the utc zone so change the zone to utc. If dt in UTC just use calendar.timegm.

def datetime_to_epoch(d1):

    # create 1,1,1970 in same timezone as d1
    d2 = datetime(1970, 1, 1, tzinfo=d1.tzinfo)
    time_delta = d1 - d2
    ts = int(time_delta.total_seconds())
    return ts


def epoch_to_datetime_string(ts, tz_name="UTC"):
    x_timezone = timezone(tz_name)
    d1 = datetime.fromtimestamp(ts, x_timezone)
    x = d1.strftime("%d %B %Y %H:%M:%S")
    return x

This class will cover your needs, you can pass the variable into ConvertUnixToDatetime & call which function you want it to operate based off.

from datetime import datetime
import time

class ConvertUnixToDatetime:
    def __init__(self, date):
        self.date = date

    # Convert unix to date object
    def convert_unix(self):
        unix = self.date

        # Check if unix is a string or int & proceeds with correct conversion
        if type(unix).__name__ == 'str':
            unix = int(unix[0:10])
        else:
            unix = int(str(unix)[0:10])

        date = datetime.utcfromtimestamp(unix).strftime('%Y-%m-%d %H:%M:%S')

        return date

    # Convert date to unix object
    def convert_date(self):
        date = self.date

        # Check if datetime object or raise ValueError
        if type(date).__name__ == 'datetime':
            unixtime = int(time.mktime(date.timetuple()))
        else:
            raise ValueError('You are trying to pass a None Datetime object')
        return type(unixtime).__name__, unixtime


if __name__ == '__main__':

    # Test Date
    date_test = ConvertUnixToDatetime(datetime.today())
    date_test = date_test.convert_date()
    print(date_test)

    # Test Unix
    unix_test = ConvertUnixToDatetime(date_test[1])
    print(unix_test.convert_unix())