Representing ambiguity in datetime?

[Terry Hancock]

What do you do when a date or time is
incompletely specified? ISTM, that as it is, there is no
formal way to store this --- you have to guess, and there's
no way to indicate that the guess is different from solid
information. As a result, I have sometimes had to abandon
datetime, even though it seemed like the logical choice for
representing data.

E.g. I might have information like "this paper was published
in May 1997". There's no way to write that with datetime,
is there? Even if I just use the "date" object instead of
datetime, I still have to actually specify something like
May 1, 1997 --- fabricating data, which is frequently
undesireable (later on, I might find information saying that
it was actually published May 23, 1997 and I might want
to update the earlier one, or simply evaluate them as
"equal" since they are, to within the precision given ---
for example, I might be trying to decide that two database
entries are really duplicate references to the same paper).

I know that this is somewhat theoretically stated, but I
have run into to concrete problems along the lines of
the above.

I'd say this is analogous to how you might use "None"
rather than "0" to represent an integer if you don't know
it's value (rather than knowing that it is zero). ISTM, you
ought to be able to specify a date as, e.g.:

d = datetime.date(2 005, 5, None)

I realize there might be some complexity with deciding
how to handle datestamp math, but as this situation
occurs frequently in real life, it seems like it shouldn't
be avoided.

How do other people deal with this kind of problem?

Cheers,
Terry

--
Terry Hancock ( hancock at anansispacework s.com )
Anansi Spaceworks http://www.anansispaceworks.com

[John Machin]

On Tue, 17 May 2005 17:38:30 -0500, Terry Hancock
<ha*****@anansi spaceworks.com> wrote:

What do you do when a date or time is
incompletely specified? ISTM, that as it is, there is no
formal way to store this --- you have to guess, and there's
no way to indicate that the guess is different from solid
information. As a result, I have sometimes had to abandon
datetime, even though it seemed like the logical choice for
representing data.

E.g. I might have information like "this paper was published
in May 1997". There's no way to write that with datetime,
is there? Even if I just use the "date" object instead of
datetime, I still have to actually specify something like
May 1, 1997 --- fabricating data, which is frequently
undesireable (later on, I might find information saying that
it was actually published May 23, 1997 and I might want
to update the earlier one, or simply evaluate them as
"equal" since they are, to within the precision given ---
for example, I might be trying to decide that two database
entries are really duplicate references to the same paper).

I know that this is somewhat theoretically stated, but I
have run into to concrete problems along the lines of
the above.

I'd say this is analogous to how you might use "None"
rather than "0" to represent an integer if you don't know
it's value (rather than knowing that it is zero). ISTM, you
ought to be able to specify a date as, e.g.:

d = datetime.date(2 005, 5, None)

I realize there might be some complexity with deciding
how to handle datestamp math, but as this situation
occurs frequently in real life, it seems like it shouldn't
be avoided.

How do other people deal with this kind of problem?

Mostly, badly :-(

Real-life example: due to war-time disruption etc, in some countries
it is common enough to find that the date of birth of someone born in
the 1940s is not known precisely. E.g. on the Hong Kong identity card,
it is possible to find only the year and month of birth, and sometimes
even only the year. Depending on the purpose, legislation and
convention will take the first day of the vague period or the last day
when a calculation is required. Badly == entering into a database the
"exact" date that was used for the purpose du jour, with no indication
that the source was vague. Consequently a person can have DOB recorded
as 1945-01-01 on one database and 1945-12-31 on another.

Suggested approach in Python (sketch): Don't try to get the datetime
module to solve the problem. Define a fuzzydate class. Internal
representation: I'd suggest earliest possible date and latest possible
date. That way you have valid date instances for doing date
arithmetic. May have different constructors depending on how the
incoming vagueness is specified.

HTH,
John

[Ron Adam]

John Machin wrote:

On Tue, 17 May 2005 17:38:30 -0500, Terry Hancock
<ha*****@anansi spaceworks.com> wrote:

What do you do when a date or time is
incompletel y specified? ISTM, that as it is, there is no
formal way to store this --- you have to guess, and there's
no way to indicate that the guess is different from solid
information . As a result, I have sometimes had to abandon
datetime, even though it seemed like the logical choice for
representin g data.

E.g. I might have information like "this paper was published
in May 1997". There's no way to write that with datetime,
is there? Even if I just use the "date" object instead of
datetime, I still have to actually specify something like
May 1, 1997 --- fabricating data, which is frequently
undesireabl e (later on, I might find information saying that
it was actually published May 23, 1997 and I might want
to update the earlier one, or simply evaluate them as
"equal" since they are, to within the precision given ---
for example, I might be trying to decide that two database
entries are really duplicate references to the same paper).

I know that this is somewhat theoretically stated, but I
have run into to concrete problems along the lines of
the above.

I'd say this is analogous to how you might use "None"
rather than "0" to represent an integer if you don't know
it's value (rather than knowing that it is zero). ISTM, you
ought to be able to specify a date as, e.g.:

d = datetime.date(2 005, 5, None)

I realize there might be some complexity with deciding
how to handle datestamp math, but as this situation
occurs frequently in real life, it seems like it shouldn't
be avoided.

How do other people deal with this kind of problem?

Mostly, badly :-(

Real-life example: due to war-time disruption etc, in some countries
it is common enough to find that the date of birth of someone born in
the 1940s is not known precisely. E.g. on the Hong Kong identity card,
it is possible to find only the year and month of birth, and sometimes
even only the year. Depending on the purpose, legislation and
convention will take the first day of the vague period or the last day
when a calculation is required. Badly == entering into a database the
"exact" date that was used for the purpose du jour, with no indication
that the source was vague. Consequently a person can have DOB recorded
as 1945-01-01 on one database and 1945-12-31 on another.

Suggested approach in Python (sketch): Don't try to get the datetime
module to solve the problem. Define a fuzzydate class. Internal
representation: I'd suggest earliest possible date and latest possible
date. That way you have valid date instances for doing date
arithmetic. May have different constructors depending on how the
incoming vagueness is specified.

HTH,
John

This is a very common problem in genealogy research as well as other
sciences that deal with history, such as geology, geography, and archeology.

I agree that some standard way of dealing with fuzzy dates would be a
good thing. I think looking at how others do it would be the way to
start...

A google search found the following reference buried in a long reference
page on mysql.

http://www.dreamlink.net/mysql/manual_Functions.html
The reason the ranges for the month and day specifiers begin

with zero is that MySQL allows incomplete dates such as
'2004-00-00' to be stored as of MySQL 3.23.
So it seems using 0's for the missing day or month may be how to do it.

Cheers,
_Ron

[Ivan Van Laningham]

Hi All--

Ron Adam wrote:

John Machin wrote:

On Tue, 17 May 2005 17:38:30 -0500, Terry Hancock
<ha*****@anansi spaceworks.com> wrote:

What do you do when a date or time is
incompletel y specified?

The reason the ranges for the month and day specifiers begin

with zero is that MySQL allows incomplete dates such as
'2004-00-00' to be stored as of MySQL 3.23.

So it seems using 0's for the missing day or month may be how to do it.

This is somewhat the approach I took in order to allow users to specify
an incomplete Mayan date in order to list possibilities. But instead of
0 (which is a valid entry in most Mayan date components), I used None.
The web version can be found at

http://www.pauahtun.org/Calendar/tools.html (the "Search for Matching
Dates" button)

The paper describing the incomplete Mayan date tool is at:
http://www.pauahtun.org/python_vuh.html

Metta,
Ivan
----------------------------------------------
Ivan Van Laningham
God N Locomotive Works
http://www.andi-holmes.com/
http://www.foretec.com/python/worksh...oceedings.html
Army Signal Corps: Cu Chi, Class of '70
Author: Teach Yourself Python in 24 Hours

[Peter Hansen]

Terry Hancock wrote:

What do you do when a date or time is
incompletely specified?

Doesn't the answer to this pretty much entirely depend on how you are
going to make use of the information? What are your use cases?

[Andrew Dalke]

Ron Adam wrote:

This is a very common problem in genealogy research as well as other
sciences that deal with history, such as geology, geography, and archeology. .. So it seems using 0's for the missing day or month may be how to do it.

Except of course humans like to make things more complicated than that.
Some journals are published quarterly so an edition might be "Jan-Mar".
Some countries refer to week numbers, so an event might be in "week 12".

I offer no suggestions as to how to handle these cases.

Andrew
da***@dalkescie ntific.com

[Dan Christensen]

Ron Adam <rr*@ronadam.co m> writes:

So it seems using 0's for the missing day or month may be how to do it.

This doesn't allow more specific amounts of ambiguity. I suggest
either a pair of dates, which represent the earliest and latest that
the event could have been (and are equal if there is no ambiguity),
or a date plus a number of days of uncertainty, i.e.
21 June 2005 +- 5 days.

Dan