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

On 1 mai, 18:15, Vadim Tropashko <vadimtro_inva..._at_yahoo.com> wrote:
> On May 1, 12:26 am, Cimode <cim..._at_hotmail.com> wrote:
>
> > My belief (and hope) would be that using interpolation
> > could represent a method more *systematic* (and therefore more easy to
> > program in a dbms) to handle missing information.
>
> One more observation. Interpolation, regression
>
> http://en.wikipedia.org/wiki/Regression_analysis
>
> , prediction, and tons of other stuff from infamous AI arena are all
> technically a join. Consider a "learning" set
>
> X Y
> -----
> 1 1
> 2 4
> 3 9
>
> and a set of unknown values
>
> X
> --
> 4
> 5
> 7
>
> for which the system is requiered to predict the Y values. "Obviously"
> the answer in this rather unsophisticated example is
>
> X Y
> -----
> 1 1
> 2 4
> 3 9
> 4 16
> 5 25
> 7 49
>
> The left outer join between the learning set and the set of unknowns
> is the mother of all prediction methods that just guesses all the
> unknown Y values to the NULLs:
>
> X Y
> -----
> 1 1
> 2 4
> 3 9
> 4 NULL
> 5 NULL
> 7 NULL
>
> Therefore, prediction operation is technically some sort of a join.

Interesting. I think more of some kind of cross intersect. I wish there would some kind of zero (instead of NULL) for solving the equation. (still digging on it) Received on Tue May 01 2007 - 20:25:43 CEST