fast-parse-time

fast-parse-time extracts calendar dates and relative time expressions from text. The calendar-date functions also accept a list of strings and can use an explicit date range to infer missing years. Parsing runs locally without network calls or a language model.

Install

Python 3.10 or later is required.

python -m pip install fast-parse-time

Find dates in a list

extract_date_strings returns the date text in the order it appears. Each list entry can contain a date, a sentence with several dates, or unrelated text. Matches retain their spelling and internal whitespace. Repeated dates remain in the result.

from fast_parse_time import extract_date_strings

texts = ['Term: 3/28/2011 to 4/22/2011', 'Mar 28', 'Apr 22', 'No classes']

assert extract_date_strings(texts) == [
    '3/28/2011', '4/22/2011', 'Mar 28', 'Apr 22',
]

Parse dates and infer missing years

parse_date_strings returns a record for each date occurrence. When at least two distinct full dates have unambiguous interpretations, the earliest and latest define an inclusive range. A date without a year is resolved only when that range supports a single calendar date.

from datetime import date
from fast_parse_time import parse_date_strings

matches = parse_date_strings(['3/28/2011', '4/22/2011', 'Mar 28', 'Apr 22'])

assert [match.date for match in matches] == [
    date(2011, 3, 28), date(2011, 4, 22),
    date(2011, 3, 28), date(2011, 4, 22),
]

The same inference works when the range crosses a year boundary. Both calendar functions accept a whole string as well as a list.

from datetime import date
from fast_parse_time import parse_date_strings

matches = parse_date_strings(
    'Term runs from 11/28/2025 to 4/14/2026; meetings on Nov 22 and March 13'
)

assert matches[-1].date == date(2026, 3, 13)
assert matches[-2].text == 'Nov 22'
assert matches[-2].year is None

March 13 occurs once inside that range, in 2026. November 22 falls outside it in both years, so its year remains unknown. The parser keeps the occurrence in the result rather than substituting the current year.

Use date_range=(start, end) to supply your own bounds. If several dates remain possible, .date is None and .candidates contains the alternatives. The calendar-date guide covers numeric date order, leap days, and unresolved results.

Extract explicit dates and relative expressions

parse_dates returns explicit date text with classification labels, along with relative expressions such as 5 days ago. Use parse_date_strings when you need calendar dates as Python date objects.

from datetime import datetime
from fast_parse_time import parse_dates

result = parse_dates('Meeting on 04/08/2024 about issues from 5 days ago')

assert result.explicit_dates[0].text == '04/08/2024'
assert result.explicit_dates[0].date_type == 'FULL_EXPLICIT_DATE'
assert result.relative_times[0].to_datetime(datetime(2026, 1, 1)) == datetime(2025, 12, 27)

You can also request relative offsets directly. Past references produce negative offsets; future references produce positive offsets.

from datetime import timedelta
from fast_parse_time import resolve_to_timedelta

assert resolve_to_timedelta('5 days ago') == [timedelta(days=-5)]
assert resolve_to_timedelta('half an hour ago') == [timedelta(minutes=-30)]
assert resolve_to_timedelta('tomorrow') == [timedelta(days=1)]

Choose a function

Interpreting results

Numeric dates can have more than one interpretation. parse_date_strings preserves that ambiguity by default; use date_order='mdy' or 'dmy' when the source's convention is known. A classification such as FULL_EXPLICIT_DATE identifies a complete date expression without selecting its month/day order.

Keep unrelated periods in separate batches, or supply explicit bounds. Relative month and year offsets use 30-day months and 365-day years. Pass a fixed reference when resolving offsets if the result needs to be reproducible.

Documentation

• API reference lists the public functions and their return values.
• Calendar dates and year inference explains batch parsing and ambiguity.
• Function examples covers extraction, filtering, and resolution.
• Result types describes the fields returned by each API.
• Development covers local tests and distribution builds.

License and author

The library is available under the MIT license. Craig Trim maintains it: craigtrim@gmail.com.

Report an issue.