We Keep Coding

How to convert time string with very precise time measurements to date objects using strptime()? [duplicate]

I am able to parse strings containing date/time with time.strptime
>>> import time
>>> time.strptime('30/03/09 16:31:32', '%d/%m/%y %H:%M:%S')
(2009, 3, 30, 16, 31, 32, 0, 89, -1)
How can I parse a time string that contains milliseconds?
>>> time.strptime('30/03/09 16:31:32.123', '%d/%m/%y %H:%M:%S')
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/lib/python2.5/_strptime.py", line 333, in strptime
data_string[found.end():])
ValueError: unconverted data remains: .123

Python 2.6 added a new strftime/strptime macro %f. The docs are a bit misleading as they only mention microseconds, but %f actually parses any decimal fraction of seconds with up to 6 digits, meaning it also works for milliseconds or even centiseconds or deciseconds.
time.strptime('30/03/09 16:31:32.123', '%d/%m/%y %H:%M:%S.%f')
However, time.struct_time doesn't actually store milliseconds/microseconds. You're better off using datetime, like this:
>>> from datetime import datetime
>>> a = datetime.strptime('30/03/09 16:31:32.123', '%d/%m/%y %H:%M:%S.%f')
>>> a.microsecond
123000
As you can see, .123 is correctly interpreted as 123 000 microseconds.

I know this is an older question but I'm still using Python 2.4.3 and I needed to find a better way of converting the string of data to a datetime.
The solution if datetime doesn't support %f and without needing a try/except is:
(dt, mSecs) = row[5].strip().split(".")
dt = datetime.datetime(*time.strptime(dt, "%Y-%m-%d %H:%M:%S")[0:6])
mSeconds = datetime.timedelta(microseconds = int(mSecs))
fullDateTime = dt + mSeconds
This works for the input string "2010-10-06 09:42:52.266000"

To give the code that nstehr's answer refers to (from its source):
def timeparse(t, format):
"""Parse a time string that might contain fractions of a second.
Fractional seconds are supported using a fragile, miserable hack.
Given a time string like '02:03:04.234234' and a format string of
'%H:%M:%S', time.strptime() will raise a ValueError with this
message: 'unconverted data remains: .234234'. If %S is in the
format string and the ValueError matches as above, a datetime
object will be created from the part that matches and the
microseconds in the time string.
"""
try:
return datetime.datetime(*time.strptime(t, format)[0:6]).time()
except ValueError, msg:
if "%S" in format:
msg = str(msg)
mat = re.match(r"unconverted data remains:"
" \.([0-9]{1,6})$", msg)
if mat is not None:
# fractional seconds are present - this is the style
# used by datetime's isoformat() method
frac = "." + mat.group(1)
t = t[:-len(frac)]
t = datetime.datetime(*time.strptime(t, format)[0:6])
microsecond = int(float(frac)*1e6)
return t.replace(microsecond=microsecond)
else:
mat = re.match(r"unconverted data remains:"
" \,([0-9]{3,3})$", msg)
if mat is not None:
# fractional seconds are present - this is the style
# used by the logging module
frac = "." + mat.group(1)
t = t[:-len(frac)]
t = datetime.datetime(*time.strptime(t, format)[0:6])
microsecond = int(float(frac)*1e6)
return t.replace(microsecond=microsecond)
raise

DNS answer above is actually incorrect. The SO is asking about milliseconds but the answer is for microseconds. Unfortunately, Python`s doesn't have a directive for milliseconds, just microseconds (see doc), but you can workaround it by appending three zeros at the end of the string and parsing the string as microseconds, something like:
datetime.strptime(time_str + '000', '%d/%m/%y %H:%M:%S.%f')
where time_str is formatted like 30/03/09 16:31:32.123.
Hope this helps.

My first thought was to try passing it '30/03/09 16:31:32.123' (with a period instead of a colon between the seconds and the milliseconds.) But that didn't work. A quick glance at the docs indicates that fractional seconds are ignored in any case...
Ah, version differences. This was reported as a bug and now in 2.6+ you can use "%S.%f" to parse it.

from python mailing lists: parsing millisecond thread. There is a function posted there that seems to get the job done, although as mentioned in the author's comments it is kind of a hack. It uses regular expressions to handle the exception that gets raised, and then does some calculations.
You could also try do the regular expressions and calculations up front, before passing it to strptime.

For python 2 i did this
print ( time.strftime("%H:%M:%S", time.localtime(time.time())) + "." + str(time.time()).split(".",1)[1])
it prints time "%H:%M:%S" , splits the time.time() to two substrings (before and after the .) xxxxxxx.xx and since .xx are my milliseconds i add the second substring to my "%H:%M:%S"
hope that makes sense :)
Example output:
13:31:21.72
Blink 01
13:31:21.81
END OF BLINK 01
13:31:26.3
Blink 01
13:31:26.39
END OF BLINK 01
13:31:34.65
Starting Lane 01

Python 3 How to format to yyyy-mm-ddThh:mm:ssZ

I'm new to Python and I cannot for the life of me find my specific answer online. I need to format a timestamp to this exact format to include 'T', 'Z' and no sub or miliseconds like this yyyy-mm-ddThh:mm:ssZ i.e. 2019-03-06T11:22:00Z. There's lots of stuff on parsing this format but nothing about formatting this way. The only way I have nearly got it to work involves sub-seconds which I do not need. I've tried using arrow and reading their documentation but unable to get anything to work. Any help would be appreciated.

Try datetime library
import datetime
output_date = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%SZ")
print(output_date)
For more information, refer to the Python Documentation.

Be careful. Just be cause a date can be formatted to look like UTC, doesn't mean it's accurate.
In ISO 8601, 'Z' is meant to designate "zulu time" or UTC ('+00:00'). While local times are typically designated by their offset from UTC. Even worse, these offsets can change throughout a year due to Daylight Saving Time (DST).
So unless you live in England in the winter or Iceland in the summer, chances are, you aren't lucky enough to be working with UTC locally, and your timestamps will be completely wrong.
Python3.8
from datetime import datetime, timezone
# a naive datetime representing local time
naive_dt = datetime.now()
# incorrect, local (MST) time made to look like UTC (very, very bad)
>>> naive_dt.strftime("%Y-%m-%dT%H:%M:%SZ")
'2020-08-27T20:57:54Z' # actual UTC == '2020-08-28T02:57:54Z'
# so we'll need an aware datetime (taking your timezone into consideration)
# NOTE: I imagine this works with DST, but I haven't verified
aware_dt = naive_dt.astimezone()
# correct, ISO-8601 (but not UTC)
>>> aware_dt.isoformat(timespec='seconds')
'2020-08-27T20:57:54-06:00'
# lets get the time in UTC
utc_dt = aware_dt.astimezone(timezone.utc)
# correct, ISO-8601 and UTC (but not in UTC format)
>>> utc_dt.isoformat(timespec='seconds')
'2020-08-28T02:57:54+00:00'
# correct, UTC format (this is what you asked for)
>>> date_str = utc_dt.isoformat(timespec='seconds')
>>> date_str.replace('+00:00', 'Z')
'2020-08-28T02:57:54Z'
# Perfect UTC format
>>> date_str = utc_dt.isoformat(timespec='milliseconds')
>>> date_str.replace('+00:00', 'Z')
'2020-08-28T02:57:54.640Z'
I just wanted to illustrate some things above, there are much simpler ways:
from datetime import datetime, timezone
def utcformat(dt, timespec='milliseconds'):
"""convert datetime to string in UTC format (YYYY-mm-ddTHH:MM:SS.mmmZ)"""
iso_str = dt.astimezone(timezone.utc).isoformat('T', timespec)
return iso_str.replace('+00:00', 'Z')
def fromutcformat(utc_str, tz=None):
iso_str = utc_str.replace('Z', '+00:00')
return datetime.fromisoformat(iso_str).astimezone(tz)
now = datetime.now(tz=timezone.utc)
# default with milliseconds ('2020-08-28T02:57:54.640Z')
print(utcformat(now))
# without milliseconds ('2020-08-28T02:57:54Z')
print(utcformat(now, timespec='seconds'))
>>> utc_str1 = '2020-08-28T04:35:35.455Z'
>>> dt = fromutcformat(utc_string)
>>> utc_str2 = utcformat(dt)
>>> utc_str1 == utc_str2
True
# it even converts naive local datetimes correctly (as of Python 3.8)
>>> now = datetime.now()
>>> utc_string = utcformat(now)
>>> converted = fromutcformat(utc_string)
>>> now.astimezone() - converted
timedelta(microseconds=997)

Thanks to skaul05 I managed to get the code I needed, it's
date = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%SZ")
print(date)

With f strings, you can shorten it down to:
from datetime import datetime
f'{datetime.now():%Y-%m-%dT%H:%M:%SZ}'
Credits go to How do I turn a python datetime into a string, with readable format date?.

Python convert string to datetime for comparison to datetime object

I have a string lfile with a datetime in it (type(lfile) gives <type 'str'>) and a Python datetime object wfile. Here is the code:
import os, datetime
lfile = '2005-08-22_11:05:45.000000000'
time_w = os.path.getmtime('{}\\{}.py' .format('C:\Temp_Readouts\RtFyar','TempReads.csv'))
wfile = datetime.datetime.fromtimestamp(time_w)
wfile contains this 2006-11-30 19:08:06.531328 and repr(wfile) gives:
datetime.datetime(2006, 11, 30, 19, 8, 6, 531328)
Problem:
I need to:
convert lfile into a Python datetime object
compare lfile to wfile and determine which datetime is more recent
For 1.:
I am only able to get a partial solution using strptime as per here. Here is what I tried:
lfile = datetime.datetime.strptime(linx_file_dtime, '%Y-%m-%d_%H:%M:%S')
The output is:
`ValueError: unconverted data remains: .000`
Question 1
It seems that strptime() cannot handle the nano seconds. How do I tell strptime() to ignore the last 3 zeros?
For 2.:
When I use type(wfile) I get <type 'datetime.datetime'>. If both wfile and lfile are Python datetime objects (i.e. if step 1. is successful), then would this work?:
if wtime < ltime:
print 'Linux file created after Windows file'
else:
print 'Windows file created after Linux file'
Question 2
Or is there some other way in which Python can compare datetime objects to determine which of the two occurred after the other?

Question 1
Python handles microseconds, not nano seconds. You can strip the last three characters of the time to convert it to microseconds and then add .%f to the end:
lfile = datetime.datetime.strptime(linx_file_dtime[:-3], '%Y-%m-%d_%H:%M:%S.%f')
Question 2
Yes, comparison works:
if wtime < ltime:
...

That's right, strptime() does not handle nanoseconds. The accepted answer in the question that you linked to offers an option: strip off the last 3 digits and then parse with .%f appended to the format string.
Another option is to use dateutil.parser.parse():
>>> from dateutil.parser import parse
>>> parse('2005-08-22_11:05:45.123456789', fuzzy=True)
datetime.datetime(2005, 8, 22, 11, 5, 45, 123456)
fuzzy=True is required to overlook the unsupported underscore between date and time components. Because datetime objects do not support nanoseconds, the last 3 digits vanish, leaving microsecond accuracy.

Is there any proper method to convert ISO8601(with tzone) to milliseconds on python? [duplicate]

This question already has answers here:
How do I parse an ISO 8601-formatted date?
(29 answers)
Closed 8 years ago.
The community reviewed whether to reopen this question last month and left it closed:
Original close reason(s) were not resolved
I'm getting a datetime string in a format like "2009-05-28T16:15:00" (this is ISO 8601, I believe). One hackish option seems to be to parse the string using time.strptime and passing the first six elements of the tuple into the datetime constructor, like:
datetime.datetime(*time.strptime("2007-03-04T21:08:12", "%Y-%m-%dT%H:%M:%S")[:6])
I haven't been able to find a "cleaner" way of doing this. Is there one?

I prefer using the dateutil library for timezone handling and generally solid date parsing. If you were to get an ISO 8601 string like: 2010-05-08T23:41:54.000Z you'd have a fun time parsing that with strptime, especially if you didn't know up front whether or not the timezone was included. pyiso8601 has a couple of issues (check their tracker) that I ran into during my usage and it hasn't been updated in a few years. dateutil, by contrast, has been active and worked for me:
from dateutil import parser
yourdate = parser.parse(datestring)

Since Python 3.7 and no external libraries, you can use the fromisoformat function from the datetime module:
datetime.datetime.fromisoformat('2019-01-04T16:41:24+02:00')
Python 2 doesn't support the %z format specifier, so it's best to explicitly use Zulu time everywhere if possible:
datetime.datetime.strptime("2007-03-04T21:08:12Z", "%Y-%m-%dT%H:%M:%SZ")

Because ISO 8601 allows many variations of optional colons and dashes being present, basically CCYY-MM-DDThh:mm:ss[Z|(+|-)hh:mm]. If you want to use strptime, you need to strip out those variations first.
The goal is to generate a UTC datetime object.
If you just want a basic case that work for UTC with the Z suffix like 2016-06-29T19:36:29.3453Z:
datetime.datetime.strptime(timestamp.translate(None, ':-'), "%Y%m%dT%H%M%S.%fZ")
If you want to handle timezone offsets like 2016-06-29T19:36:29.3453-0400 or 2008-09-03T20:56:35.450686+05:00 use the following. These will convert all variations into something without variable delimiters like 20080903T205635.450686+0500 making it more consistent/easier to parse.
import re
# This regex removes all colons and all
# dashes EXCEPT for the dash indicating + or - utc offset for the timezone
conformed_timestamp = re.sub(r"[:]|([-](?!((\d{2}[:]\d{2})|(\d{4}))$))", '', timestamp)
datetime.datetime.strptime(conformed_timestamp, "%Y%m%dT%H%M%S.%f%z" )
If your system does not support the %z strptime directive (you see something like ValueError: 'z' is a bad directive in format '%Y%m%dT%H%M%S.%f%z') then you need to manually offset the time from Z (UTC). Note %z may not work on your system in Python versions < 3 as it depended on the C library support which varies across system/Python build type (i.e., Jython, Cython, etc.).
import re
import datetime
# This regex removes all colons and all
# dashes EXCEPT for the dash indicating + or - utc offset for the timezone
conformed_timestamp = re.sub(r"[:]|([-](?!((\d{2}[:]\d{2})|(\d{4}))$))", '', timestamp)
# Split on the offset to remove it. Use a capture group to keep the delimiter
split_timestamp = re.split(r"([+|-])",conformed_timestamp)
main_timestamp = split_timestamp[0]
if len(split_timestamp) == 3:
sign = split_timestamp[1]
offset = split_timestamp[2]
else:
sign = None
offset = None
# Generate the datetime object without the offset at UTC time
output_datetime = datetime.datetime.strptime(main_timestamp +"Z", "%Y%m%dT%H%M%S.%fZ" )
if offset:
# Create timedelta based on offset
offset_delta = datetime.timedelta(hours=int(sign+offset[:-2]), minutes=int(sign+offset[-2:]))
# Offset datetime with timedelta
output_datetime = output_datetime + offset_delta

Arrow looks promising for this:
>>> import arrow
>>> arrow.get('2014-11-13T14:53:18.694072+00:00').datetime
datetime.datetime(2014, 11, 13, 14, 53, 18, 694072, tzinfo=tzoffset(None, 0))
Arrow is a Python library that provides a sensible, intelligent way of creating, manipulating, formatting and converting dates and times. Arrow is simple, lightweight and heavily inspired by moment.js and requests.

You should keep an eye on the timezone information, as you might get into trouble when comparing non-tz-aware datetimes with tz-aware ones.
It's probably the best to always make them tz-aware (even if only as UTC), unless you really know why it wouldn't be of any use to do so.
#-----------------------------------------------
import datetime
import pytz
import dateutil.parser
#-----------------------------------------------
utc = pytz.utc
BERLIN = pytz.timezone('Europe/Berlin')
#-----------------------------------------------
def to_iso8601(when=None, tz=BERLIN):
if not when:
when = datetime.datetime.now(tz)
if not when.tzinfo:
when = tz.localize(when)
_when = when.strftime("%Y-%m-%dT%H:%M:%S.%f%z")
return _when[:-8] + _when[-5:] # Remove microseconds
#-----------------------------------------------
def from_iso8601(when=None, tz=BERLIN):
_when = dateutil.parser.parse(when)
if not _when.tzinfo:
_when = tz.localize(_when)
return _when
#-----------------------------------------------

I haven't tried it yet, but pyiso8601 promises to support this.

import datetime, time
def convert_enddate_to_seconds(self, ts):
"""Takes ISO 8601 format(string) and converts into epoch time."""
dt = datetime.datetime.strptime(ts[:-7],'%Y-%m-%dT%H:%M:%S.%f')+\
datetime.timedelta(hours=int(ts[-5:-3]),
minutes=int(ts[-2:]))*int(ts[-6:-5]+'1')
seconds = time.mktime(dt.timetuple()) + dt.microsecond/1000000.0
return seconds
This also includes the milliseconds and time zone.
If the time is '2012-09-30T15:31:50.262-08:00', this will convert into epoch time.
>>> import datetime, time
>>> ts = '2012-09-30T15:31:50.262-08:00'
>>> dt = datetime.datetime.strptime(ts[:-7],'%Y-%m-%dT%H:%M:%S.%f')+ datetime.timedelta(hours=int(ts[-5:-3]), minutes=int(ts[-2:]))*int(ts[-6:-5]+'1')
>>> seconds = time.mktime(dt.timetuple()) + dt.microsecond/1000000.0
>>> seconds
1348990310.26

Both ways:
Epoch to ISO time:
isoTime = time.strftime('%Y-%m-%dT%H:%M:%SZ', time.gmtime(epochTime))
ISO time to Epoch:
epochTime = time.mktime(time.strptime(isoTime, '%Y-%m-%dT%H:%M:%SZ'))

Isodate seems to have the most complete support.

aniso8601 should handle this. It also understands timezones, Python 2 and Python 3, and it has a reasonable coverage of the rest of ISO 8601, should you ever need it.
import aniso8601
aniso8601.parse_datetime('2007-03-04T21:08:12')

Here is a super simple way to do these kind of conversions.
No parsing, or extra libraries required.
It is clean, simple, and fast.
import datetime
import time
################################################
#
# Takes the time (in seconds),
# and returns a string of the time in ISO8601 format.
# Note: Timezone is UTC
#
################################################
def TimeToISO8601(seconds):
strKv = datetime.datetime.fromtimestamp(seconds).strftime('%Y-%m-%d')
strKv = strKv + "T"
strKv = strKv + datetime.datetime.fromtimestamp(seconds).strftime('%H:%M:%S')
strKv = strKv +"Z"
return strKv
################################################
#
# Takes a string of the time in ISO8601 format,
# and returns the time (in seconds).
# Note: Timezone is UTC
#
################################################
def ISO8601ToTime(strISOTime):
K1 = 0
K2 = 9999999999
K3 = 0
counter = 0
while counter < 95:
K3 = (K1 + K2) / 2
strK4 = TimeToISO8601(K3)
if strK4 < strISOTime:
K1 = K3
if strK4 > strISOTime:
K2 = K3
counter = counter + 1
return K3
################################################
#
# Takes a string of the time in ISO8601 (UTC) format,
# and returns a python DateTime object.
# Note: returned value is your local time zone.
#
################################################
def ISO8601ToDateTime(strISOTime):
return time.gmtime(ISO8601ToTime(strISOTime))
#To test:
Test = "2014-09-27T12:05:06.9876"
print ("The test value is: " + Test)
Ans = ISO8601ToTime(Test)
print ("The answer in seconds is: " + str(Ans))
print ("And a Python datetime object is: " + str(ISO8601ToDateTime(Test)))

datetime: Round/trim number of digits in microseconds

Currently I am logging stuff and I am using my own formatter with a custom formatTime():
def formatTime(self, _record, _datefmt):
t = datetime.datetime.now()
return t.strftime('%Y-%m-%d %H:%M:%S.%f')
My issue is that the microseconds, %f, are six digits. Is there anyway to spit out less digits, like the first three digits of the microseconds?

The simplest way would be to use slicing to just chop off the last three digits of the microseconds:
def format_time():
t = datetime.datetime.now()
s = t.strftime('%Y-%m-%d %H:%M:%S.%f')
return s[:-3]
I strongly recommend just chopping. I once wrote some logging code that rounded the timestamps rather than chopping, and I found it actually kind of confusing when the rounding changed the last digit. There was timed code that stopped running at a certain timestamp yet there were log events with that timestamp due to the rounding. Simpler and more predictable to just chop.
If you want to actually round the number rather than just chopping, it's a little more work but not horrible:
def format_time():
t = datetime.datetime.now()
s = t.strftime('%Y-%m-%d %H:%M:%S.%f')
head = s[:-7] # everything up to the '.'
tail = s[-7:] # the '.' and the 6 digits after it
f = float(tail)
temp = "{:.03f}".format(f) # for Python 2.x: temp = "%.3f" % f
new_tail = temp[1:] # temp[0] is always '0'; get rid of it
return head + new_tail
Obviously you can simplify the above with fewer variables; I just wanted it to be very easy to follow.

As of Python 3.6 the language has this feature built in:
def format_time():
t = datetime.datetime.now()
s = t.isoformat(timespec='milliseconds')
return s

This method should always return a timestamp that looks exactly like this (with or without the timezone depending on whether the input dt object contains one):
2016-08-05T18:18:54.776+0000
It takes a datetime object as input (which you can produce with datetime.datetime.now()). To get the time zone like in my example output you'll need to import pytz and pass datetime.datetime.now(pytz.utc).
import pytz, datetime
time_format(datetime.datetime.now(pytz.utc))
def time_format(dt):
return "%s:%.3f%s" % (
dt.strftime('%Y-%m-%dT%H:%M'),
float("%.3f" % (dt.second + dt.microsecond / 1e6)),
dt.strftime('%z')
)
I noticed that some of the other methods above would omit the trailing zero if there was one (e.g. 0.870 became 0.87) and this was causing problems for the parser I was feeding these timestamps into. This method does not have that problem.

An easy solution that should work in all cases:
def format_time():
t = datetime.datetime.now()
if t.microsecond % 1000 >= 500: # check if there will be rounding up
t = t + datetime.timedelta(milliseconds=1) # manually round up
return t.strftime('%Y-%m-%d %H:%M:%S.%f')[:-3]
Basically you do manual rounding on the date object itself first, then you can safely trim the microseconds.

Edit: As some pointed out in the comments below, the rounding of this solution (and the one above) introduces problems when the microsecond value reaches 999500, as 999.5 is rounded to 1000 (overflow).
Short of reimplementing strftime to support the format we want (the potential overflow caused by the rounding would need to be propagated up to seconds, then minutes, etc.), it is much simpler to just truncate to the first 3 digits as outlined in the accepted answer, or using something like:
'{:03}'.format(int(999999/1000))
-- Original answer preserved below --
In my case, I was trying to format a datestamp with milliseconds formatted as 'ddd'. The solution I ended up using to get milliseconds was to use the microsecond attribute of the datetime object, divide it by 1000.0, pad it with zeros if necessary, and round it with format. It looks like this:
'{:03.0f}'.format(datetime.now().microsecond / 1000.0)
# Produces: '033', '499', etc.

You can subtract the current datetime from the microseconds.
d = datetime.datetime.now()
current_time = d - datetime.timedelta(microseconds=d.microsecond)
This will turn 2021-05-14 16:11:21.916229 into 2021-05-14 16:11:21

This method allows flexible precision and will consume the entire microsecond value if you specify too great a precision.
def formatTime(self, _record, _datefmt, precision=3):
dt = datetime.datetime.now()
us = str(dt.microsecond)
f = us[:precision] if len(us) > precision else us
return "%d-%d-%d %d:%d:%d.%d" % (dt.year, dt.month, dt.day, dt.hour, dt.minute, dt.second, int(f))
This method implements rounding to 3 decimal places:
import datetime
from decimal import *
def formatTime(self, _record, _datefmt, precision='0.001'):
dt = datetime.datetime.now()
seconds = float("%d.%d" % (dt.second, dt.microsecond))
return "%d-%d-%d %d:%d:%s" % (dt.year, dt.month, dt.day, dt.hour, dt.minute,
float(Decimal(seconds).quantize(Decimal(precision), rounding=ROUND_HALF_UP)))
I avoided using the strftime method purposely because I would prefer not to modify a fully serialized datetime object without revalidating it. This way also shows the date internals in case you want to modify it further.
In the rounding example, note that the precision is string-based for the Decimal module.

Here is my solution using regexp:
import re
# Capture 6 digits after dot in a group.
regexp = re.compile(r'\.(\d{6})')
def to_splunk_iso(dt):
"""Converts the datetime object to Splunk isoformat string."""
# 6-digits string.
microseconds = regexp.search(dt.isoformat()).group(1)
return regexp.sub('.%d' % round(float(microseconds) / 1000), dt.isoformat())

Fixing the proposed solution based on Pablojim Comments:
from datetime import datetime
dt = datetime.now()
dt_round_microsec = round(dt.microsecond/1000) #number of zeroes to round
dt = dt.replace(microsecond=dt_round_microsec)

If once want to get the day of the week (i.e, 'Sunday)' along with the result, then by slicing '[:-3]' will not work. At that time you may go with,
dt = datetime.datetime.now()
print("{}.{:03d} {}".format(dt.strftime('%Y-%m-%d %I:%M:%S'), dt.microsecond//1000, dt.strftime("%A")))
#Output: '2019-05-05 03:11:22.211 Sunday'
%H - for 24 Hour format
%I - for 12 Hour format
Thanks,

Adding my two cents here as this method will allow you to write your microsecond format as you would a float in c-style. It takes advantage that they both use %f.
import datetime
import re
def format_datetime(date, format):
"""Format a ``datetime`` object with microsecond precision.
Pass your microsecond as you would format a c-string float.
e.g "%.3f"
Args:
date (datetime.datetime): You input ``datetime`` obj.
format (str): Your strftime format string.
Returns:
str: Your formatted datetime string.
"""
# We need to check if formatted_str contains "%.xf" (x = a number)
float_format = r"(%\.\d+f)"
has_float_format = re.search(float_format, format)
if has_float_format:
# make microseconds be decimal place. Might be a better way to do this
microseconds = date.microsecond
while int(microseconds): # quit once it's 0
microseconds /= 10
ms_str = has_float_format.group(1) % microseconds
format = re.sub(float_format, ms_str[2:], format)
return date.strftime(format)
print(datetime.datetime.now(), "%H:%M:%S.%.3f")
# '17:58:54.424'