Re: A new proof of the superiority of set oriented approaches: numerical/time serie linear interpolation

[Snipped]
> > Quite frankly, I am not interested that much into a terminology debate
> > but rather into *how interpolation could eventually serve as a way to
> > handle missing information*. So far, Vadim has brought some
> > interesting insight through the addition of a theorical regression
> > method implemented in SQL (I will try to adapt it as soon as I find
> > sometime). I am afraid the discussions about *design quality*, *SQL
> > critiscism* are in fact off topic.
>
> I apologize for my part in hijacking your thread.
That's fine. You are only minorily responsible for the off topic trend on the thread.

> One thing you might want to pursue that would interest me is "what kinds of
> data lend themselves to interpolation, regression, or any other kind of
> smoothing, and what kinds do not."
I believe this is the kind of comments and question I was hoping to trigger.

The purpose of this thread I believe to refine thought about the handling of *some* missing data instead of design issues.

> If we know that on June 10, a certain person's name was Mary Smith, and on
> September 3, her name was Mary Jones, interpolation is not going to help
> us figure out her name on sugust 17.
Correct. See below.

> Is it only numerical data that lend themselves to the kind of approximation
> and/or inference that you are exploring?
Yes. To be more concise numerical and date time series. I am somehow convinced that it would be more practical to implement a dbms using the method of interpolation as it would be using existing methods of decomposition. So I am trying to identify some methods on a per case basis that would viable for a safe implementation. It seems to me that in case of mathematically inspired series, a dbms would benefit from probabilistic method for handling missing data.

Regards... Received on Thu May 03 2007 - 17:20:06 CEST