Date: Sat, 29 Nov 2014 20:01:27 -0700 (MST)
From: David G Johnston <david.g.johnston@gmail.com>
To: pgsql-sql@postgresql.org
Message-ID: <1417316487139-5828676.post@n5.nabble.com>
In-Reply-To: 
 <CALN462YFf+Nu+8HGd6+7kYDj5Gbb9XEsasZ5d3c9Y=FUOaryug@mail.gmail.com>
References: 
 <CALN462YFf+Nu+8HGd6+7kYDj5Gbb9XEsasZ5d3c9Y=FUOaryug@mail.gmail.com>
Subject: Re: regr_slope function with auto creation of X column
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-sql-owner@postgresql.org

Jason Aleksi wrote
> --Storing 30 day regression slopes into table (DOES NOT WORK)
> INSERT INTO historical_data_regr_slope (department_id, date,
> regr_slope30sales) (
>    SELECT historical_data.department_id, historical_data.date,
>           regr_slope(row_number(), historical_data.salesDollarK) OVER
> (PARTITION BY historical_data.department_id ORDER BY historical_data.date
> DESC ROWS BETWEEN 1 PRECEDING AND 29 FOLLOWING) AS regr_slope30sales
>    FROM historical_data
>    GROUP BY historical_data.department_id, historical_data.date,
> historical_data.salesDollarK
>    ORDER BY historical_data.department_id, historical_data.date DESC
> )
> 
> Any suggestions on how to auto-create the regr_slope X column?

You shoud be able to use a subquery to first generate the relevant row
numbers and then in the outer query apply the regr_slope function.

You could also try (theory here - the documentation should be improved in
this area) two applications of the OVER clause.

regr_expr( row_number() over (...), sales ) over (...)

You should probably define the window in the main body and refer to it by
name if you attempt this.  I honestly have no idea if it will work but the
syntax you used before is defined as invalid because nothing can come
between the function and the OVER part; which negates the possibility of
using a single OVER to cover two functions.

I would suggest you not intermix window and group by until you get the
window working.  Then put that into a cte/with and run the group by
separately - you might need to put the group by in the cte and the window in
the main query.  That said I haven't fully contemplated what it is you are
attempting to calculate.  Typically moving averages are not going to require
a group by clause...you just need to add a where clause that can filter out
the first N records where the number of input rows is less than N.

David J.


--
View this message in context: http://postgresql.nabble.com/regr-slope-function-with-auto-creation-of-X-column-tp5828673p5828676.html
Sent from the PostgreSQL - sql mailing list archive at Nabble.com.


-- 
Sent via pgsql-sql mailing list (pgsql-sql@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-sql