DateDetective
Introduction
DateDetective is a Python package that takes a machine learning approach to identifying the format of date strings. This tool is useful for many applications like web scraping where the amount of formats used to represent dates is many and there is no need for 100% accuracy.
Compatible date formats
DateDetective's model is trained to predict what combination of Python datetime module format codes would make up a given string representation of a date. Currently the model can identify the following format codes:
Format Code | Description | Examples |
---|
%d | Day of the month as zero-padded decimal number | 01, 02, ..., 30, 31 |
%B | Month as full text name | January, February, March, ..., December |
%b | Month as abrieviated text name | Jan, Feb, Mar, ..., Dec |
%m | Month as a zero-padded decimal number | 01, 02, 03, ..., 12 |
%Y | Year with century as decimal number | 1832, 1996, 2002, 2024 |
%H | Hours as zero-padded decimal number (24 hour clock) | 00, 01, 02, ..., 22, 23, 24 |
%I | Hours as zero-padded decimal number (12 hour clock) | 01, 02, 03, ..., 10, 11, 12 |
%M | Minutes as zero-padded decimal number | 00, 01, 02, ..., 58, 59, 60 |
%S | Seconds as zero-padded decimal number | 00, 01, 02, ..., 58, 59, 60 |
%f | Microsecond as decimal number, zero-padded to six digits | 000000, 000001, ..., 999999 |
%p | AM or PM | AM, PM |
%Z | Time zone name as text | UTC, GMT, EAT, EDT |
%z | Time zone as UTC offset decimal number | +0000, -1200, +1000 |
Installation
Firstly you will need to ensure that the version of PyTorch that is best for you is installed in the Python environment you are using. If possible use CUDA as this should increase date detection speed. Currently PyTorch's website has an install command finder at:
https://pytorch.org/get-started/locally/
After PyTorch is installed you can install DateDetective via Pip.
pip install DateDetective
Usage
Import and Initialise
from datedetective import DateDetective
dd = DateDetective()
By default DateDetective will use CUDA cores on your GPU (if available) for some of the calculations. If you do not want to use CUDA then initialise DateFinder like this:
df = DateDetective(useCuda=False)
Generate datetime module format string from date string
>>>dd.get_format("30/12/2023 12:52:23")
'%d/%m/%Y %H:%M:%S'
Create a datetime object string from date string
>>>dd.get_datetime("30/12/2023 12:52:23")
datetime.datetime(2023, 12, 30, 12, 52, 23)
Generate datetime module format string from list of date strings with same format
DateDetective is more accurate if you have multiple date strings written in the same format, the following examples all benefit from this increased accuracy.
>>>date_str_list = ["31/12/1997", "20/01/2015", "01/01/2003", "01/12/2010", "23/08/1954", "15/05/2016", "30/03/2022", "11/06/2007"]
>>>dd.get_list_format(date_str_list)
"%d/%m/%Y"
Convert all date strings in a list to datetime objects
>>>date_str_list = ["31/12/1997", "20/01/2015", "01/01/2003", "23/08/1954", "30/03/2022"]
>>>dd.get_list_datetime(date_str_list)
[datetime.datetime(1997, 12, 31, 0, 0, 0), datetime.datetime(2015, 1, 20, 0, 0, 0), datetime.datetime(2003, 1, 1, 0, 0, 0), datetime.datetime(1954, 8, 23, 0, 0, 0), datetime.datetime(2022, 3, 30, 0, 0, 0)]
Generate format of date strings that are contained in a list of dictionaries
>>>dict_list = [{"name": "Alison", "date_of_birth": "31/12/1997"},
{"name": "Rory", "date_of_birth": "20/01/2015"},
{"name": "Charlotte", "date_of_birth": "01/01/2003"},
{"name": "Jo", "city": "London"}
{"name": "Geoff", "date_of_birth": "23/08/1954"},
{"name": "Rob", "date_of_birth": "30/03/2022"}]
>>>dd.get_dict_list_format(dict_list, "date_of_birth")
"%d/%m/%Y"
When using a function that takes lists of dictionaries you must specify the key for each dictionary that stores the date strings that DateDetective will predict the format for.
As seen in the example above and in following example, not all dictionaries in the list provided need to contain the date string key (i.e. "Jo"). DateDetective will skip these dictionaries.
Convert date strings in a list of dictionaries to datetime objects
>>>dict_list = [{"name": "Alison", "date_of_birth": "31/12/1997"},
{"name": "Rory", "date_of_birth": "20/01/2015"},
{"name": "Charlotte", "date_of_birth": "01/01/2003"},
{"name": "Jo", "city": "London"}
{"name": "Geoff", "date_of_birth": "23/08/1954"},
{"name": "Rob", "date_of_birth": "30/03/2022"}]
>>>dd.get_dict_list_datetime(dict_list, "date_of_birth")
[{"name": "Alison", "date_of_birth": datetime.datetime(1997, 12, 31, 0, 0, 0)},
{"name": "Rory", "date_of_birth": datetime.datetime(2015, 1, 20, 0, 0, 0)},
{"name": "Charlotte", "date_of_birth": datetime.datetime(2003, 1, 1, 0, 0, 0)},
{"name": "Jo", "city": "London"}
{"name": "Geoff", "date_of_birth": datetime.datetime(1954, 8, 23, 0, 0, 0)},
{"name": "Rob", "date_of_birth": datetime.datetime(2022, 3, 30, 0, 0, 0)}]
If you set retain_date_str to True then the returned list of dictionaries will also contain the original date strings. They will be stored under the date key with "_original" concatenated on the end.
>>>dict_list = [{"name": "Alison", "date_of_birth": "31/12/1997"},
...
{"name": "Rob", "date_of_birth": "30/03/2022"}]
>>>dd.get_dict_list_datetime(dict_list, "date_of_birth", retain_date_str=True)
[{"name": "Alison", "date_of_birth": datetime.datetime(1997, 12, 31, 0, 0, 0), "date_of_birth_original": "31/12/1997"},
...
{"name": "Rob", "date_of_birth": datetime.datetime(2022, 3, 30, 0, 0, 0), "date_of_birth_original": "30/03/2022"}]
It's important to remember that although DateFinder is usually accurate it sometimes makes mistakes.
How it was trained
Take a look at my repo that is dedicated to the training of the DateDetective models for more information: DateDetective Training GitHub Repo
License
Apache License 2.0. See LICENSE file.
Contact
Rob Salethorn - rob@salethorn.com
Repo link - https://github.com/RSalethorn/DateDetective