Different where condition for each column - sql

Question

Is there a way to write query like this in SQL Server, without using select two times and then join?
select trans_date, datepart(HOUR,trans_time) as hour,
(datepart(MINUTE,trans_time)/30)*30 as minute,
case
when paper_number = 11111/*paperA*/
then sum(t1.price*t1.amount)/SUM(t1.amount)*100
end as avgA,
case
when paper_number = 22222/*PaperB*/
then sum(t1.price*t1.amount)/SUM(t1.amount)*100
end as avgB
from dbo.transactions t1
where trans_date = '2006-01-01' and (paper_number = 11111 or paper_number = 22222)
group by trans_date, datepart(HOUR,trans_time), datepart(MINUTE,trans_time)/30
order by hour, minute
SQL Server asks me to add paper_number to group by, and returns nulls when I do so
trans_date hour minute avgA avgB
2006-01-01 9 30 1802.57199725463 NULL
2006-01-01 9 30 NULL 169125.886524823
2006-01-01 10 0 1804.04742534103 NULL
2006-01-01 10 0 NULL 169077.777777778
2006-01-01 10 30 1806.18773535637 NULL
2006-01-01 10 30 NULL 170274.550381867
2006-01-01 11 0 1804.43466045433 NULL
2006-01-01 11 0 NULL 170743.4
2006-01-01 11 30 1807.04532012137 NULL
2006-01-01 11 30 NULL 171307.00280112

Answer 1

Try:
with cte as
(select trans_date,
datepart(HOUR,trans_time) as hour,
(datepart(MINUTE,trans_time)/30)*30 as minute,
sum(case when paper_number = 11111/*paperA*/
then t1.price*t1.amount else 0 end) as wtdSumA,
sum(case when paper_number = 11111/*paperA*/
then t1.amount else 0 end) as amtSumA,
sum(case when paper_number = 22222/*PaperB*/
then t1.price*t1.amount else 0 end) as wtdSumB,
sum(case when paper_number = 22222/*PaperB*/
then t1.amount else 0 end) as amtSumB
from dbo.transactions t1
where trans_date = '2006-01-01'
group by trans_date, datepart(HOUR,trans_time), datepart(MINUTE,trans_time)/30)
select trans_date, hour, minute,
case amtSumA when 0 then 0 else 100 * wtdSumA / amtSumA end as avgA,
case amtSumB when 0 then 0 else 100 * wtdSumB / amtSumB end as avgB
from cte
order by hour, minute
(SQLFiddle here)
You can derive this without the CTE, like so:
select trans_date,
datepart(HOUR,trans_time) as hour,
(datepart(MINUTE,trans_time)/30)*30 as minute,
case sum(case when paper_number = 11111/*paperA*/ then t1.amount else 0 end)
when 0 then 0
else 100 * sum(case when paper_number = 11111 then t1.price*t1.amount else 0 end)
/ sum(case when paper_number = 11111 then t1.amount else 0 end) end as avgA,
case sum(case when paper_number = 22222/*paperA*/ then t1.amount else 0 end)
when 0 then 0
else 100 * sum(case when paper_number = 22222 then t1.price*t1.amount else 0 end)
/ sum(case when paper_number = 22222 then t1.amount else 0 end) end as avgB
from dbo.transactions t1
where trans_date = '2006-01-01'
group by trans_date, datepart(HOUR,trans_time), datepart(MINUTE,trans_time)/30
order by 1,2,3

Answer 2

Use SUM() function on the entire CASE expression
select trans_date, datepart(HOUR,trans_time) as hour, (datepart(MINUTE,trans_time)/30)*30 as minute,
sum(case when paper_number = 11111/*paperA*/ then t1.price*t1.amount end) * 1.00
/ sum(case when paper_number = 11111/*paperA*/ then t1.amount end) * 100 as avgA,
sum(case when paper_number = 22222/*PaperB*/ then t1.price*t1.amount end) * 1.00
/ sum(case when paper_number = 22222/*paperB*/ then t1.amount end) * 100 as avgB
from dbo.transactions t1
where trans_date = '2006-01-01'
group by trans_date, datepart(HOUR,trans_time), datepart(MINUTE,trans_time)/30
order by hour, minute
Demo on SQLFiddle

Answer 3

You could also try using UNPIVOT and PIVOT like below:
WITH prepared AS (
SELECT
trans_date,
trans_time = DATEADD(MINUTE, DATEDIFF(MINUTE, '00:00', trans_time) / 30 * 30, CAST('00:00' AS time)),
paper_number,
total = price * amount,
amount
FROM transactions
),
unpivoted AS (
SELECT
trans_date,
trans_time,
attribute = attribute + CAST(paper_number AS varchar(10)),
value
FROM prepared
UNPIVOT (value FOR attribute IN (total, amount)) u
),
pivoted AS (
SELECT
trans_date,
trans_time,
avgA = total11111 * 100 / amount11111,
avgB = total22222 * 100 / amount22222
FROM unpivoted
PIVOT (
SUM(value) FOR attribute IN (total11111, amount11111, total22222, amount22222)
) p
)
SELECT *
FROM pivoted
;
As an attempt at explaining how the above query works, below is a description of transformations that the original dataset undergoes in the course of the query's execution, using the following example:
trans_date trans_time paper_number price amount
---------- ---------- ------------ ----- ------
2013-04-09 11:12:35 11111 10 15
2013-04-09 11:13:01 22222 24 10
2013-04-09 11:28:44 11111 12 5
2013-04-09 11:36:20 22222 20 11
The prepared CTE produces the following column set:
trans_date trans_time paper_number total amount
---------- ---------- ------------ ----- ------
2013-04-09 11:00:00 11111 150 15
2013-04-09 11:00:00 22222 240 10
2013-04-09 11:00:00 11111 60 5
2013-04-09 11:30:00 22222 220 11
where trans_time is the original trans_time rounded down to the nearest half-hour and total is price multiplied by amount.
The unpivoted CTE unpivots the total and amount values to produce attribute and value:
trans_date trans_time paper_number attribute value
---------- ---------- ------------ --------- -----
2013-04-09 11:00:00 11111 total 150
2013-04-09 11:00:00 11111 amount 15
2013-04-09 11:00:00 22222 total 240
2013-04-09 11:00:00 22222 amount 10
2013-04-09 11:00:00 11111 total 60
2013-04-09 11:00:00 11111 amount 5
2013-04-09 11:30:00 22222 total 220
2013-04-09 11:30:00 22222 amount 11
Then paper_number is combined with attribute to form a single column, also called attribute:
trans_date trans_time attribute value
---------- ---------- ----------- -----
2013-04-09 11:00:00 total11111 150
2013-04-09 11:00:00 amount11111 15
2013-04-09 11:00:00 total22222 240
2013-04-09 11:00:00 amount22222 10
2013-04-09 11:00:00 total11111 60
2013-04-09 11:00:00 amount11111 5
2013-04-09 11:30:00 total22222 220
2013-04-09 11:30:00 amount22222 11
Finally, the pivoted CTE pivots the value data back aggregating them along the way with SUM() and using the attribute values for column names:
trans_date trans_time total11111 amount11111 total22222 amount22222
---------- ---------- ---------- ----------- ---------- -----------
2013-04-09 11:00:00 210 20 240 10
2013-04-09 11:30:00 NULL NULL 220 11
The pivoted values are then additionally processed (every totalNNN is multiplied by 100 and divided by the corresponding amountNNN) to form the final output:
trans_date trans_time avgA avgB
---------- ---------- ---- ----
2013-04-09 11:00:00 1050 2400
2013-04-09 11:30:00 NULL 2000
There's a couple of issues that may need to be addressed:
If price and amount are different data types, the total and amount may end up different data types as well. For UNPIVOT, it is mandatory that the values being unpivoted are of exactly the same type, and so you'll need to add an explicit conversion of total and amount to some common type, possibly one which would prevent data/precision loss. That would could be done in the prepared CTE like this (assuming the common type to be decimal(10,2)):
total = CAST(price * amount AS decimal(10,2)),
amount = CAST(amount AS decimal(10,2))
If aggregated amounts may ever end up 0, you'll need to account for the division by 0 issue. One way to do that could be to substitute the 0 amount with NULL, which would make the result of the division NULL as well. Applying ISNULL or COALESCE to that result would allow you to transform it to some default value, 0 for instance. So, change this bit in the pivoted CTE:
avgA = ISNULL(total11111 * 100 / NULLIF(amount11111, 0), 0),
avgB = ISNULL(total22222 * 100 / NULLIF(amount22222, 0), 0)