Обсуждение: user defined aggregate with multiple arguments

I have a need to create some statistical functions in PostgreSQL that
will take multiple "column" arguments and return multiple values.

A good example would be a least squares fit of a line through a set of
(x,y) points. In this case, I would want the function to accept two
"columns" of data (i.e. x data and y data) as arguments and return two
parameters: the slope and intercept of a line.

The slope is given by
   [sum(x^2)*sum(y) - sum(x)*sum*x*y)] / [M*sum(x^2) -(sum(x))^2]
It's not *that* complicated, but it's not that easy to remember either.
And if you're already doing a lot of work to tease out the x and y
points from various tables, it simplifies your query if you can do a
function call.

I can't tell from what have read about user defined functions and user
defined aggregates whether this kind of function is possible.

Bill Eaton

 > I have a need to create some statistical functions in PostgreSQL that
will take multiple
 > "column" arguments and return multiple values.
 >
 > A good example would be a least squares fit of a line through a set
of (x,y) points.
 > In this case, I would want the function to accept two "columns" of
data (i.e. x data and
 > y data) as arguments and return two parameters: the slope and
intercept of a line.
 >
 > The slope is given by   [sum(x^2)*sum(y) - sum(x)*sum*x*y)] /
[M*sum(x^2) -(sum(x))^2]
 >
 > It's not *that* complicated, but it's not that easy to remember
either. And if you're already
 > doing a lot of work to tease out the x and y points from various
tables, it simplifies your query
 > if you can do a function call.
 >
 > I can't tell from what have read about user defined functions and
user defined aggregates whether
 > this kind of function is possible.
 >
 > Bill Eaton

I just notice that multiple inputs to aggregates are allowed in the
upcoming 8.2 release. This gives me a great incentive to play with the
beta and upgrade from 8.0.

One question remains: how about multiple outputs? Can I have a ROW as a
return value -- i.e. something like SLOPE and INTERCEPT in the original
example?

-Bill Eaton

On Wed, 11 Oct 2006, Bill Eaton wrote:

>
> I just notice that multiple inputs to aggregates are allowed in the upcoming
> 8.2 release. This gives me a great incentive to play with the beta and
> upgrade from 8.0.
>
> One question remains: how about multiple outputs? Can I have a ROW as a
> return value -- i.e. something like SLOPE and INTERCEPT in the original
> example?
>

First, you don't have to write these functions by yourself. The functions
for least squares fit are defined in the standart, and they will be also
in 8.2.
http://momjian.us/main/writings/pgsql/sgml/functions-aggregate.html

But, in any way, if you want your aggregate to return several values, you
can always return them as the elements in the array.

Regards,
     Sergey

*******************************************************************
Sergey E. Koposov
Max Planck Institute for Astronomy/Sternberg Astronomical Institute
Tel: +49-6221-528-349
Web: http://lnfm1.sai.msu.ru/~math
E-mail: math@sai.msu.ru

> On Wed, 11 Oct 2006, Bill Eaton wrote:
>>
>> I just notice that multiple inputs to aggregates are allowed in the
>> upcoming 8.2 release. This gives me a great incentive to play with
>> the beta and upgrade from 8.0.
>>
>> One question remains: how about multiple outputs? Can I have a ROW as
>> a return value -- i.e. something like SLOPE and INTERCEPT in the
>> original example?
>>
> Sergey E. Koposov wrote:
> First, you don't have to write these functions by yourself. The
> functions for least squares fit are defined in the standart, and they
> will be also in 8.2.
> http://momjian.us/main/writings/pgsql/sgml/functions-aggregate.html
>
> But, in any way, if you want your aggregate to return several values,
> you can always return them as the elements in the array.
Actually, I was wanting to do higher order polynomial fits. But it's
cool that there are some new functions. So you suggest I use an array as
a return type. Hmm. Never did much with arrays before. So that means a
"row" or "setof" is not an option?

-Bill