timezonefinder 1.5.3

Python library to look up timezone from lat / long offline. Improved version of "pytzwhere".

This is a fast and lightweight python project to lookup the corresponding timezone for a given lat/lng on earth entirely offline.

This project is derived from and has been successfully tested against pytzwhere (github), but aims to provide improved performance and usability.

It is also similar to django-geo-timezones

The underlying timezone data is based on work done by Eric Muller.

Timezones at sea and Antarctica are not yet supported (because somewhat special rules apply there).

Dependencies

(python, math, struct, os)

numpy

Optional:

Numba and its Requirements

This is only for precompiling the time critical algorithms. When you only look up a few points once in a while, the compilation time is probably outweighing the benefits. When using certain_timezone_at() and especially closest_timezone_at() however, I highly recommend using numba (see speed comparison below)! The amount of shortcuts used in the .bin is also only optimized for the use with numba.

Installation

(install the dependencies)

in your terminal simply:

pip install timezonefinder

(you might need to run this command as administrator)

Usage

Basics:

from timezonefinder import TimezoneFinder

tf = TimezoneFinder()

for testing if numba is being used: (if the import of the optimized algorithms worked)

print(TimezoneFinder.using_numba())
# this is a static method returning True or False

fast algorithm:

# point = (longitude, latitude)
point = (13.358, 52.5061)
print( tf.timezone_at(*point) )
# = Europe/Berlin

To make sure a point is really inside a timezone (slower):

print( tf.certain_timezone_at(*point) )
# = Europe/Berlin

To find the closest timezone (slow):

# only use this when the point is not inside a polygon!
# this checks all the polygons within +-1 degree lng and +-1 degree lat
point = (12.773955, 55.578595)
print( tf.closest_timezone_at(*point) )
# = Europe/Copenhagens

To increase search radius even more (very slow, use ``numba``!):

# this checks all the polygons within +-3 degree lng and +-3 degree lat
# I recommend only slowly increasing the search radius
# keep in mind that x degrees lat are not the same distance apart than x degree lng!
print( tf.closest_timezone_at(lng=point[0],lat=point[1],delta_degree=3) )
# = Europe/Copenhagens

(to make sure you really got the closest timezone increase the search radius until you get a result. then increase the radius once more and take this result.)

Further application:

To maximize the chances of getting a result in a Django view it might look like:

def find_timezone(request, lat, lng):
    lat = float(lat)
    lng = float(lng)

    try:
        timezone_name = tf.timezone_at(lng, lat)
        if timezone_name is None:
            timezone_name = tf.closest_timezone_at(lng, lat)
            # maybe even increase the search radius when it is still None

    except ValueError:
        # the coordinates were out of bounds
        # {handle error}

    # ... do something with timezone_name ...

To get an aware datetime object from the timezone name:

# first pip install pytz
from pytz import timezone, utc
from pytz.exceptions import UnknownTimeZoneError

# tzinfo has to be None (means naive)
naive_datetime = YOUR_NAIVE_DATETIME

try:
    tz = timezone(timezone_name)
    aware_datetime = naive_datetime.replace(tzinfo=tz)
    aware_datetime_in_utc = aware_datetime.astimezone(utc)

    naive_datetime_as_utc_converted_to_tz = tz.localize(naive_datetime)

except UnknownTimeZoneError:
    # ... handle the error ...

also see the pytz Doc.

Using the conversion tool:

Place the tz_world.csv from tzwhere in one folder with the file_converter.py and run it as a script. It converts the .csv in a new .csv and transforms this file into the needed .bin

Place this .bin in your timezonfinder folder (overwriting the old file) to make it being used.

Please note: Neither the tests nor the file_converter.py are optimized or really beautiful. Sorry for that.

Comparison to pytzwhere

In comparison to pytzwhere I managed to speed up the queries by more than 100 times (s. test results below). Initialisation time and memory usage are also significanlty reduced, while my algorithm yields the same results. In some cases pytzwhere even does not find anything and timezonefinder does, for example when only one timezone is close to the point.

Similarities:

• results

• data being used

Differences:

• the data is now stored in a memory friendly 18MB .bin and needed data is directly being read on the fly (instead of reading and converting the 76MB .csv -mostly floats stored as strings!- into memory every time a class is created).

• precomputed shortcuts are stored in the .bin to quickly look up which polygons have to be checked (instead of creating the shortcuts on every startup)

• optimized algorithms

• introduced proximity algorithm

• use of numba for speeding things up much further.

test results from the latest version *:

test correctness:
Results:
[point, target, tzwere is correct, timezonefinder is correct]
(-60.968888, -3.442172) America/Manaus True True
(14.1315716, 2.99999) Africa/Douala True True
(-106.1706459, 23.7891123) America/Mazatlan True True
(33, -84) uninhabited True True
(103.7069307, 1.3150701) Asia/Singapore True True
(-71.9996885, -52.7868679) America/Santiago True True
(-4.8663325, 40.0663485) Europe/Madrid True True
(-152.4617352, 62.3415036) America/Anchorage True True
(-44.7402611, 70.2989263) America/Godthab True True
(12.9125913, 50.8291834) Europe/Berlin True True
(37.0720767, 55.74929) Europe/Moscow True True
(14.1315716, 0.2350623) Africa/Brazzaville True True

testing 10000 realistic points
MISMATCHES:
(-110.46557383479337, 35.860783819335666) America/Phoenix America/Denver (not counted, see issue section)
(28.33811173592602, -30.053783637444724) Africa/Johannesburg Africa/Maseru (not counted, see issue section)

testing 10000 random points
MISMATCHES:
(27.86670722464703, -29.135850729733704) Africa/Johannesburg Africa/Maseru (not counted, see issue section)

in 20000 tries 0 mismatches were made
fail percentage is: 0.0


testing certain_timezone_at():

testing 10000 realistic points
MISMATCHES:
(-110.46557383479337, 35.860783819335666) America/Phoenix America/Denver (not counted, see issue section)
(28.33811173592602, -30.053783637444724) Africa/Johannesburg Africa/Maseru (not counted, see issue section)

testing 10000 random points
MISMATCHES:

in 20000 tries 0 mismatches were made
fail percentage is: 0.0

w/ Numba:
TIMES for  10000 realistic queries:
tzwhere: 0:02:55.985141
timezonefinder: 0:00:00.905828
194.28 times faster

TIMES for  10000 random queries:
tzwhere: 0:01:29.427567
timezonefinder: 0:00:00.604325
147.98 times faster


w/o Numba:
TIMES for  1000 realistic queries:
tzwhere: 0:00:17.819268
timezonefinder: 0:00:03.269472
5.45 times faster


TIMES for  1000 random queries:
tzwhere: 0:00:09.189154
timezonefinder: 0:00:01.748470
5.26 times faster


Startup times:
tzwhere: 0:00:08.302153
timezonefinder: 0:00:00.008768
946.87 times faster

* Version: 1.5.3 System: MacBookPro 2,4GHz i5 4GB RAM SSD tzwhere: numpy=True, shapely=False

**mismatch: pytzwhere finds something and then timezonefinder finds something else

***realistic queries: just points within a timezone (= pytzwhere yields result)

****random queries: random points on earth

Speed Impact of Numba

TIMES for 1000 realistic queries***:

timezone_at():
wo/ numa: 0:00:01.017575
w/ numa: 0:00:00.289854
3.51 times faster

certain_timezone_at():
wo/ numa:   0:00:05.445209
w/ numa: 0:00:00.290441
14.92 times faster

closest_timezone_at():
(delta_degree=1)
wo/ numa: 0:02:32.666238
w/ numa: 0:00:02.688353
40.2 times faster

(this is not included in my tests)

Known Issues

I ran tests for approx. 5M points and this are the mistakes I found:

All points in Lesotho are counted to the ‘Africa/Johannesburg’ timezone instead of ‘Africa/Maseru’. I am pretty sure this is because it is completely surrounded by South Africa and in the data the area of Lesotho is not excluded from this timezone.

Same for the small usbekish enclaves in Kirgisitan and some points in the Arizona Dessert (some weird rules apply here).

Those are mistakes in the data not my algorithms and in order to fix this I would need check for and then separately handle these special cases. This would not only slow down the algorithms, but also make them ugly.

Contact

This is the first public python project I did, so most certainly there is stuff I missed, things I could have optimized even further etc. That’s why, I would be really glad to get feedback on my code.

If you notice that the tz data is outdated, encounter any bugs, have suggestions, criticism, etc. feel free to open an Issue, add Pull Requests on Git or …

contact me: python at michelfe dot it

License

timezonefinder is distributed under the terms of the MIT license (see LICENSE.txt).

Changelog

Note: not mentioned versions only contain small and irrelevant changes (e.g. in the readme, setup.py…). I am new to all this, so I am often missing small things which are not really new features worth mentioning.

1.5.3 (2016-04-23)

• using 32-bit ints for storing the polygons now (instead of 64-bit): I calculated that the minimum accuracy (at the equator) is 1cm with the approach I use. Tests passed.

• Benefits: 18MB file instead of 35MB, another 10-30% speed boost (depending on your hardware)

1.5.2 (2016-04-20)

• added python 2.7.6 support: replaced strings in unpack (unsupported by python 2.7.6 or earlier) with byte strings

• timezone names are now loaded from a separate file for better modularity

1.5.1 (2016-04-18)

• added python 2.7.8+ support:

  Therefore I had to change the tests a little bit (some operations were not supported). This only affects output. I also had to replace one part of the algorithms to prevent overflow in Python 2.7

1.5.0 (2016-04-12)

• automatically using optimized algorithms now (when numba is installed)

• added TimezoneFinder.using_numba() function to check if the import worked

1.4.0 (2016-04-07)

• Added the file_converter.py to the repository: It converts the .csv from pytzwhere to another .csv and this one into the used .bin.

  Especially the shortcut computation and the boundary storage in there save a lot of reading and computation time, when deciding which timezone the coordinates are in. It will help to keep the package up to date, even when the timezone data should change in the future.