Question

I have a table that looks like this:

CREATE TABLE foobar (
                     id                     SERIAL PRIMARY KEY,
                     data_entry_date        DATE NOT NULL,
                     user_id                INTEGER NOT NULL,
                     wine_glasses_drunk     INTEGER NOT NULL,
                     whisky_shots_drunk     INTEGER NOT NULL,
                     beer_bottle_drunk      INTEGER NOT NULL
                 );

insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-01', 1, 1,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-02', 1, 4,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-03', 1, 0,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-04', 1, 1,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-05', 1, 2,1,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-07', 1, 1,2,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-08', 1, 4,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-11', 1, 1,1,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-12', 1, 1,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-13', 1, 2,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-14', 1, 1,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-15', 1, 9,3,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-16', 1, 0,4,2);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-17', 1, 0,5,3);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-18', 1, 2,2,5);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-20', 1, 1,1,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-23', 1, 1,3,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-01-24', 1, 0,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-02-01', 1, 1,1,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-02-02', 1, 2,3,4);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-02-05', 1, 1,2,2);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-02-09', 1, 0,0,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-02-10', 1, 1,1,1);
insert into foobar (data_entry_date, user_id, wine_glasses_drunk, whisky_shots_drunk, beer_bottle_drunk) VALUES ('2011-02-11', 1, 3,6,3);

I want to write a query that shows me the difference in TOTAL wine_glasses_drunk, TOTAL whisky_shots_drunk and TOTAL beer_bottles_drunk over a given period, compared to the TOTALs for the previous period.

It probably sounds more complicated than it is. If we are using a period* of 1 week == 7 days, then the query should return the difference in the totals consumed for this week, as compared to the totals consumed last week.

A slight complication is that the dates in the table are not continuous – i.e. there are some missing dates, so the query needs to find the most relevant date when determining dates for period calculations.

This is what I have so far:

-- using hard coded dates

SELECT (SUM(f1.wine_glasses_drunk) - SUM(f2.wine_glasses_drunk)) as wine_diff, 
(SUM(f1.whisky_shots_drunk) - SUM(f2.whisky_shots_drunk)) as whisky_diff, 
(SUM(f1.beer_bottle_drunk) - SUM(f2.beer_bottle_drunk)) as beer_diff 
FROM foobar f1 INNER JOIN foobar f2 ON f2.user_id=f1.user_id
WHERE f1.user_id=1 
AND f1.data_entry_date BETWEEN '2011-01-08' AND '2011-01-15'
AND f2.data_entry_date BETWEEN '2011-01-01' AND '2011-01-08'
AND f1.data_entry_date - f2.data_entry_date between 6 and 9;

The above SQL is clearly a hack (especially the f1.data_entry_date - f2.data_entry_date between 6 and 9 criteria). I checked the results in excel, and the results from the query above were (unsuprisingly) wrong.

How can I write this query – and how can I modify it so that it can deal with the non contiguous dates in the database?

I am using postgreSQl, but would prefer if possible, database agnostic (i.e. ANSI) SQL.

Answer 1

I was going to add this as an edit to my other answer, but it really is a different way of doing it so should be a separate answer.

I think I prefer the other answer that I have given but this one should work even if there are gaps in the data.

To set the parameters for the query, change the values of period_start_date and period_days in the query_params part of the with clause.

with query_params as (
  select 
    date '2011-01-01' as period_start_date,
    7 as period_days
),
summary_data as (
select
  user_id,
  (data_entry_date - period_start_date)/period_days as period_number,
  sum(wine_glasses_drunk) as wine_glasses_drunk,
  sum(whisky_shots_drunk) as whisky_shots_drunk,
  sum(beer_bottle_drunk) as beer_bottle_drunk
from foobar
  cross join query_params
group by user_id,
  (data_entry_date - period_start_date)/period_days
)
select
  user_id,
  period_number,
  period_start_date + period_number * period_days as period_start_date,
  sum(wine_glasses_drunk) as wine_glasses_drunk,
  sum(whisky_shots_drunk) as whisky_shots_drunk,
  sum(beer_bottle_drunk) as beer_bottle_drunk
from (
  -- this weeks data
  select 
    user_id,
    period_number,
    wine_glasses_drunk,
    whisky_shots_drunk,
    beer_bottle_drunk
  from summary_data
  union all
  -- last weeks data
  select 
    user_id,
    period_number + 1 as period_number,
    -wine_glasses_drunk as wine_glasses_drunk,
    -whisky_shots_drunk as whisky_shots_drunk,
    -beer_bottle_drunk as beer_bottle_drunk
  from summary_data
) a
cross join query_params
where period_number <= (select max(period_number) from summary_data)
group by 
  user_id,
  period_number,
  period_start_date + period_number * period_days
order by 1, 2

And again, a SQL Fiddle is available.