Comments on: Lag, lead and sparse data http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/ Delivered Intelligence Mon, 15 Mar 2010 14:12:52 -0400 http://wordpress.org/?v=2.8.4 hourly 1 By: Peter Scott http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/comment-page-1/#comment-4451 Peter Scott Wed, 11 Apr 2007 06:21:22 +0000 http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/#comment-4451 Alex Thanks for the correction - I should have been clearer in that I was referring to a <i>synthetic</i> row generator source such as those cited by Paweł on his blog; you are quite right I would have to have a list of available date for the join - often I use a calendar table; guess that's sticking to my Star Schema roots. Alex
Thanks for the correction – I should have been clearer in that I was referring to a synthetic row generator source such as those cited by Paweł on his blog; you are quite right I would have to have a list of available date for the join – often I use a calendar table; guess that’s sticking to my Star Schema roots.

]]> By: Alex Gorbachev http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/comment-page-1/#comment-4456 Alex Gorbachev Wed, 11 Apr 2007 01:56:36 +0000 http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/#comment-4456 <i>but another approach (Oracle 10g) is to use a partition outer join.</i> Wouldn't you still use some kind row-generator source to join with? but another approach (Oracle 10g) is to use a partition outer join.
Wouldn’t you still use some kind row-generator source to join with?

]]> By: Peter Scott http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/comment-page-1/#comment-4454 Peter Scott Tue, 10 Apr 2007 20:47:57 +0000 http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/#comment-4454 Thanks Paweł The original post was spurred by a notion that lag and lead are row based and not range based. Filling in the gaps to "put in the missing zeros" is indeed probably the way most people would go for this case - row generators are one one way (as you rightly show) but another approach (Oracle 10g) is to use a partition outer join. Thanks Paweł
The original post was spurred by a notion that lag and lead are row based and not range based.

Filling in the gaps to “put in the missing zeros” is indeed probably the way most people would go for this case – row generators are one one way (as you rightly show) but another approach (Oracle 10g) is to use a partition outer join.

]]> By: Paweł Barut http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/comment-page-1/#comment-4453 Paweł Barut Tue, 10 Apr 2007 20:29:41 +0000 http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/#comment-4453 Hope this time it fill work: <code> select sal_dt, quantity, nvl(quantity,0) - nvl(SUM(quantity) over (order by sal_dt range between interval '1' day preceding and interval '1' day preceding ),0) delta from ps_lag_1, (select (select min(sales_date) from ps_lag_1)+level-1 sal_dt from dual connect by level <= (select max(sales_date)-min(sales_date)+2 from PS_LAG_1)) where sal_dt = sales_date(+); </code> Hope this time it fill work:

select sal_dt, quantity, nvl(quantity,0) - nvl(SUM(quantity) over (order by sal_dt range between interval '1' day preceding and interval '1' day preceding ),0) delta
from ps_lag_1, (select (select min(sales_date) from ps_lag_1)+level-1 sal_dt
from dual connect by level <= (select max(sales_date)-min(sales_date)+2 from PS_LAG_1))
where sal_dt = sales_date(+);

]]> By: Paweł Barut http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/comment-page-1/#comment-4452 Paweł Barut Tue, 10 Apr 2007 20:28:06 +0000 http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/#comment-4452 It seems that code was broken in above comment. It should be: select sal_dt, quantity, nvl(quantity,0) - nvl(SUM(quantity) over (order by sal_dt range between interval '1' day preceding and interval '1' day preceding ),0) delta from ps_lag_1, (select (select min(sales_date) from ps_lag_1)+level-1 sal_dt from dual connect by level It seems that code was broken in above comment. It should be:
select sal_dt, quantity, nvl(quantity,0) – nvl(SUM(quantity) over (order by sal_dt range between interval ‘1′ day preceding and interval ‘1′ day preceding ),0) delta
from ps_lag_1,
(select (select min(sales_date) from ps_lag_1)+level-1 sal_dt
from dual connect by level

]]> By: Paweł Barut http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/comment-page-1/#comment-4455 Paweł Barut Tue, 10 Apr 2007 20:26:27 +0000 http://www.rittmanmead.com/2007/04/10/lag-lead-and-sparse-data/#comment-4455 Well, but still on 3rd April sales was 3 less than on second. I would rather use <a href="http://pbarut.blogspot.com/2006/09/row-generators.html" rel="nofollow">row generator</a> to fill in gaps. <code> select sal_dt, quantity, nvl(quantity,0) - nvl(SUM(quantity) over (order by sal_dt range between interval '1' day preceding and interval '1' day preceding ),0) delta from ps_lag_1, (select (select min(sales_date) from ps_lag_1)+level-1 sal_dt from dual connect by level that gives: <code> 07/04/01 2 2 07/04/02 3 1 07/04/03 -3 07/04/04 5 5 07/04/05 5 0 07/04/06 3 -2 07/04/07 -3 </code> Regards, Paweł Well, but still on 3rd April sales was 3 less than on second. I would rather use row generator to fill in gaps.

select sal_dt, quantity, nvl(quantity,0) - nvl(SUM(quantity) over (order by sal_dt range between interval '1' day preceding and interval '1' day preceding ),0) delta
from ps_lag_1,
(select (select min(sales_date) from ps_lag_1)+level-1 sal_dt
from dual connect by level that gives:

07/04/01 2 2
07/04/02 3 1
07/04/03 -3
07/04/04 5 5
07/04/05 5 0
07/04/06 3 -2
07/04/07 -3

Regards,
Paweł

]]>