Analytic function to retrieve a value one year ago

Question

Hello,
I'm trying to find an analytic function to get a value on another row by looking on a date with Oracle 11gR2.
I have a table with a date_id (truncated date), a flag and a measure. For each date, I have at least one row (sometimes 2), so it is gapless.
I would like to find analytic functions to show for each date :
sum of the measure for that date
sum of the measure one week ago
sum of the measure one year ago
As it is gapless I managed to do it the week doing a group by date in a subquery and using a LAG with offset set to 7 on top of it (see below).
However I'm struggling on how to do that for the data one year ago as we might have leap years. I cannot simply set the offset to 365.
Is it possible to do it with a RANGE BETWEEN window clause? I can't manage to have it working with dates.
Week :LAG with offset 7
SQL Fiddle
or
create table daily_counts
  date_id date,
  internal_flag number,
  measure1 number
insert into daily_counts values ('01-Jan-2013', 0, 8014);
insert into daily_counts values ('01-Jan-2013', 1, 2);
insert into daily_counts values ('02-Jan-2013', 0, 1300);
insert into daily_counts values ('02-Jan-2013', 1, 37);
insert into daily_counts values ('03-Jan-2013', 0, 19);
insert into daily_counts values ('03-Jan-2013', 1, 14);
insert into daily_counts values ('04-Jan-2013', 0, 3);
insert into daily_counts values ('05-Jan-2013', 0, 0);
insert into daily_counts values ('05-Jan-2013', 1, 1);
insert into daily_counts values ('06-Jan-2013', 0, 0);
insert into daily_counts values ('07-Jan-2013', 1, 3);
insert into daily_counts values ('08-Jan-2013', 0, 33);
insert into daily_counts values ('08-Jan-2013', 1, 9);
commit;
select
    date_id,
    total1,
    LAG(total1, 7) OVER(ORDER BY date_id) total_one_week_ago
  from
      select
        date_id,
        SUM(measure1) total1
      from daily_counts
      group by date_id
order by 1;
Year : no idea?
I can't give a gapless example, would be too long but if there is a solution with the date directly :
SQL Fiddle
or add this to the schema above :
insert into daily_counts values ('07-Jan-2012', 0, 11);
insert into daily_counts values ('07-Jan-2012', 1, 1);
insert into daily_counts values ('08-Jan-2012', 1, 4);
Thank you for your help.
Floyd

Answer 1

Hi,
Sorry, I;m not sure I understand the problem.
If you are certain that there is at least 1 row for every day, then you can be sure that the GROUP BY will produce exactly 1 row per day, and you can use LAG (total1, 365) just like you already use LAG (total1, 7).
Are you concerned about leap years?  That is, when the day is March 1, 2016, do you want the total_one_year_ago column to reflect March 1, 2015, which was 366 days earlier?  If that case, use
date_id - ADD_MONTHS (date_id, -12)
instead of  365.
LAG only works with an exact number, but you can use RANGE BETWEEN with other analytic functions, such as MIN or SUM:
SELECT DISTINCT
          date_id
,         SUM (measure1) OVER (PARTITION BY date_id)    AS total1
,         SUM (measure1) OVER ( ORDER BY      date_id
                                RANGE BETWEEN 7 PRECEDING
                                      AND     7 PRECEDING
                              )                       AS total1_one_week_ago
,         SUM (measure1) OVER ( ORDER BY      date_id
                                RANGE BETWEEN 365 PRECEDING
                                      AND     365 PRECEDING
                              )                       AS total1_one_year_ago
FROM      daily_counts
ORDER BY  date_id
Again, use date arithmetic instead of the hard-coded 365, if that's an issue.
As Hoek said, it really helps to post the exact results you want from the given sample data.  You're miles ahead of the people who don't even post the sample data, though.
You're right not to post hundreds of INSERT statements to get a year's data.  Here's one way to generate sample data for lots of rows at the same time:
-- Put a 0 into the table for every day in 2012
INSERT INTO daily_counts (date_id, measure1)
SELECT  DATE '2011-12-31' + LEVEL
,       0
FROM    dual
CONNECT BY LEVEL <= 366