unit_of_time
is a small package to represents time ranges through an int
, this means we can easily store these, for
example in a database. It also offers functions to determine the previous and next time range.
The package by default comes with a year, quarter, month, week and day as time units. We want to be able to convert a certain week into an int, and back. For this, we have defined the following rule:
- a time range should encode to one integer, not per se the other way around;
- the time ranges should be orderable such that we can easily order and sort items in the database; and
- it should be easy for humans to read the integers.
For this we have defined the following format:
YYYYMMDDK
Here YYYY-MM-DD is the start date of the time range, and K is the "kind" of time range. The longer a time unit, the lower its kind.
So a year for example has kind 1
, this means that we encode the year 2025 as 202501011
, the month january 2025 is encoded as 202501015
.
If we thus order the time units, we first sort on the start of the time unit, and the ordering will put longer time units first, so the "year 2025 starts 'earlier' than the quarter 2025Q1 which starts earlier than 2025-01".
The package provides some utility functions to make working with the time units more convenient.
For example when we construct a year with
from datetime import date
from unit_of_time import Year
year1958 = Year(date(1958, 3, 25))
we can go to the next and previous year with:
year1959 = year1958.next
year1957 = year1958.previous
one can also use .ancestors
and .successors
which are generators that will keep proposing previous and next time units respectively, so we can walk over the years since 1958 with:
for year in year1958.successors:
print(year)
We can also determine if a date, or another time range is fully enclosed by another one, for example:
from datetime import date
from unit_of_time import Month, Year
Month(date(1958, 3, 25)) in Year(date(1958, 3, 25)) # True
Month(date(2019, 11, 25)) in Year(date(1958, 3, 25)) # False
or check if there is overlap between two time units, especially since weeks are not always fully enclosed by the month, quarter, etc. when the week starts or ends. For example:
from datetime import date
from unit_of_time import Week, Year
Week(date(1957, 12, 31)).overlaps_with(Year(date(1958, 1, 1))) # True
since the week with 1957-12-31 starts on December 30th.
We can also check if one time unit starts before another time unit if these are of the same kind, like:
from datetime import date
from unit_of_time import Week, Week
Week(date(1957, 12, 31)) <= Week(date(1958, 3, 25))
A time unit itself is iterable: it will yield all dates contained by the time unit. For example we can get all dates of 1958Q1
with:
from datetime import date
from unit_of_time import Quarter
for dt in Quarter(date(1958, 3, 25)):
print(dt) # 1958-01-01 to 1958-03-31
we can also convert such collection to a list.
The units of time can also be shifted, for example:
from datetime import date
from unit_of_time import Day, Month, Quarter, Week, Year
Year(date(1958, 3, 25)) << 1 # Year(date(1959, 1, 1))
Quarter(date(1958, 3, 25)) >> 2 # Quarter(date(1957, 7, 1))
3 >> Month(date(1958, 3, 25)) # Month(date(1958, 6, 1))
4 << Week(date(1958, 3, 25)) # Week(date(1958, 2, 24))
Year(date(1958, 3, 25)) << -1 # Year(date(1957, 1, 1))
Quarter(date(1958, 3, 25)) >> -2 # Quarter(date(1958, 7, 1))
-3 >> Month(date(1958, 3, 25)) # Month(date(1958, 12, 1))
-4 << Week(date(1958, 3, 25)) # Week(date(1958, 4, 21))
so we can add or subtract a given number of years, quarters, months, weeks, and days from a given unit of time.
A time unit is hashable, it uses the int
representation as hash. It is also indexable, and uses the int
representation.
We can thus make a (very) long list, and work with:
from datetime import date
from unit_of_time import Day
specials_unit_of_times = [False] * 202512319
specials_unit_of_times[Day(date(1958, 3, 25))] = True
we can even use this to slice, although it probably is not very useful.
We can register a new time unit. For example, a decade with:
from datetime import date
from unit_of_time import TimeunitKind
class Decade(TimeunitKind):
kind_int = 0
formatter = "%Ys"
@classmethod
def truncate(cls, dt):
return date(dt.year - dt.year % 10, 1, 1)
@classmethod
def _next(cls, dt):
return date(dt.year + 10, 1, 1)
Subclassing TimeunitKind
will automatically register it. One has to fill in the kind_int
, which is an integer, preferrably between 0
and 9
, although one can register outside the range. If that is the case, the "kind" will take two or more digits when converting to an int.
One can also implement a formatter. This is strictly speaking not necessary, since one can also implement a .to_str()
method:
from unit_of_time import TimeunitKind
class Decade(TimeunitKind):
kind_int = 0
def to_str(cls, dt):
return dt.strftime('%Ys')
# ...
this might be useful if the formatting is more advanced than what Python's date formatter can handle.
Furthermore, one implements the .truncate(..)
class method to convert a date to the start of the date range, and the _next(..)
which returns the first date for the next decade.
With these functions, we have registered a new time unit.
The package ships with the following time units:
- 1: year;
- 3: quarter;
- 5: month;
- 7: week; and
- 9: day.
There is deliberately always one integer between the two time units, such that one can always put a customized one between any of the two.
195803259 – 201911259