To: "Ding Xiangguang" <Xiangguang@singtel.com>
Cc: <pgsql-sql@postgresql.org>
Subject: Re: view of weekly data
References: 
 <543CACD7FF1C4F4AB587CC0864E94EC603B5E31D@EMHQ05B.singtel.corp.root>
In-Reply-To: 
 <543CACD7FF1C4F4AB587CC0864E94EC603B5E31D@EMHQ05B.singtel.corp.root>
From: Greg Stark <gsstark@mit.edu>
Organization: The Emacs Conspiracy; member since 1992
Date: 18 Feb 2006 02:36:25 -0500
Message-ID: <87hd6x15l2.fsf@stark.xeocode.com>
Lines: 55
User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.4
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

"Ding Xiangguang" <Xiangguang@singtel.com> writes:

> Hi, friend,
> 
> Suppose there is table of daily transaction data with 5 fields,
> 
> time(date), open(float8), high(float8), low(float8), close(float8)
> 
> Is it possible to create a view of weekly data, i.e. open is the first
> day'open, high is the highest of the week, low is the lowest of the week, close
> is the last day's close.

low and high are easy, they're just min() and max(). so you would get
something like:

select date_trunc('week', time) as startofweek, 
       min(low) as weeklylow, 
       max(high) as weeklyhigh
  from dailydata
 group by date_trunc('week', time)


Unfortunately showing the open and close is much much harder. To get them
efficiently requires a feature set called OLAP that Postgres doesn't have and
isn't likely to get soon.

In Postgres 8.1 (and 8.0?) you could actually write some custom aggregate
functions using RECORD data type to store the earliest and latest time found
so far and the corresponding open and close to get them efficiently. Maybe
someone else would be able to show how to do that, I haven't tried it yet.

The only way to do it in standardish SQL would be with terribly inefficient
subqueries:

select date_trunc('week', time) as startofweek, 
       min(low) as weeklylow, 
       max(high) as weeklyhigh,
       (select open 
          from dailydata as d 
         where date_trunc('week',time)=date_trunc('week',dailydata.time)
         order by time asc
         limit 1
       ) as weeklyopen,
       (select close
          from dailydata as d 
         where date_trunc('week',time)=date_trunc('week',dailydata.time)
         order by time desc
         limit 1
       ) as weeklyclose
  from dailydata
 group by date_trunc('week', time)


-- 
greg