Re: Temporal database - no end date
From: V.J. Kumar <vjkmail_at_gmail.com>
Date: Fri, 19 Jan 2007 22:27:42 +0100 (CET)
Message-ID: <Xns98BDA7BB2C069vdghher_at_194.177.96.26>
>
> Le sigh.
>
> I am making a very, very narrow claim. I would appreciate it if you
> read what I wrote, and addressed your questions to it.
>
> My technical observation is that using chronons to discretize time
> implies that mathematical operations over temporal quantities lose
> information.
>
> They either don't, or they do it badly. How do you propose to
> compute
> the variance of a temporal random variable using integers?
> academic question, or corner case from obscure scientific
> applications. It's a basic question that comes up regularly in
> industrial and OR applications all the time.
>
> I'm perfectly happy to make the necessary engineering compromises
> and
> use doubles precision types in my algorithms, mindful of the joys that
> come with 'thinking about precision'. Alternatively I could-- and have
> before -- opt to use one of the quite sophisticated open source math
> packages that support arbitrarily large integers, and whatever
> rational numbers can be expressed using ( INTEGER / INTEGER ) ( a
> perfectly reasonably domain which can be implemented in most of the
> modern SQL DBMS engines).
>
> NONE of which is relevant to the question of whether my temporal
> domain is better off modelled as a sequence of discrete units, or
> points on a continium.
Date: Fri, 19 Jan 2007 22:27:42 +0100 (CET)
Message-ID: <Xns98BDA7BB2C069vdghher_at_194.177.96.26>
"DBMS_Plumber" <paul_geoffrey_brown_at_yahoo.com> wrote in news:1169239340.938970.215230_at_s34g2000cwa.googlegroups.com:
>> So, how do you model your continuous time if all you have is at best >> a subset of rational numbers (IEEE 754) ?
>
> Le sigh.
>
> I am making a very, very narrow claim. I would appreciate it if you
> read what I wrote, and addressed your questions to it.
>
> My technical observation is that using chronons to discretize time
> implies that mathematical operations over temporal quantities lose
> information.
Could you show with an example how the loss occurs ? You may be right, but let's see.
> My snarky remark was to the effect that many people on
> this list are perfectly happy to invoke the 'but the physical
> computer' dodge when it suits them, while objecting to it's use at
> other times.
>
>> >Models of time that divide the continium into >> > discrete units, and then force all intervals, aggregations and the >> > results of any operation into that model, just don't work. >> >> Really ? How come that people do it all the time by using digital >> computers that do not have real numbers ? All the computers have are >> a subset of integers and a subset of rationals ?
>
> They either don't, or they do it badly. How do you propose to
> compute
> the variance of a temporal random variable using integers?
>This is no
> academic question, or corner case from obscure scientific
> applications. It's a basic question that comes up regularly in
> industrial and OR applications all the time.
>
> I'm perfectly happy to make the necessary engineering compromises
> and
> use doubles precision types in my algorithms, mindful of the joys that
> come with 'thinking about precision'. Alternatively I could-- and have
> before -- opt to use one of the quite sophisticated open source math
> packages that support arbitrarily large integers, and whatever
> rational numbers can be expressed using ( INTEGER / INTEGER ) ( a
> perfectly reasonably domain which can be implemented in most of the
> modern SQL DBMS engines).
>
> NONE of which is relevant to the question of whether my temporal
> domain is better off modelled as a sequence of discrete units, or
> points on a continium.
Here you go again. You do not have "points on a continium" in your computer, you must use integers or rational numbers, both of which are really the same in the sense of being able to approximate the continuum.
>
>
Received on Fri Jan 19 2007 - 22:27:42 CET