oracle group by when value is zero, aggregate = 0 - sql

Question

i am trying to build a query but it's taking me too much time to resolve it.
Oracle database v18
this is my table1
Date1
tagname
Value
01/01/2021 0:01
a
2
01/01/2021 0:02
a
4
01/01/2021 0:01
b
2
01/01/2021 0:02
b
4
01/01/2021 0:01
c
2
01/01/2021 0:02
c
4
02/01/2021 0:01
a
0
02/01/2021 0:02
a
0
02/01/2021 0:01
b
2
02/01/2021 0:02
b
4
02/01/2021 0:01
c
2
02/01/2021 0:02
c
4
i am doing an average by day
select avg(value) value, tagname, to_date(date1,'dd/MM/yyyy')
from table1
group by date1, tagname
Result:
Date1
tagname
Value
01/01/2021
a
3
01/01/2021
b
3
01/01/2021
c
3
02/01/2021
a
0
02/01/2021
b
3
02/01/2021
c
3
now i need to add a new tagname
select sum(value), 'newtag' tagname
from result
where tagname= 'a' or tagname = 'b' or tagname= 'c'
group by date1
but when a=0 newtag value = 0
How could i archieve this?
example
Date1
tagname
Value
01/01/2021
a
3
01/01/2021
b
3
01/01/2021
c
3
01/01/2021
newtag
9
02/01/2021
a
0
02/01/2021
b
3
02/01/2021
c
3
02/01/2021
newtag
0
could i use case in this query?
thanks in advance
Edit: table1 have more tagnames, but only need to sum(a+b+c)

Answer 1

So, this is easy to do with UNION ALL, of course. I guess your concern is that you do not want to read through your table twice (once to calculate date/tag aggregates and again to calculate date aggregates).
Anytime you want to aggregate query results at multiple levels, you should at least consider GROUPING SETS functionality.
The trick in your case isn't the multiple-level aggregates. Rather, it is that you want the 2nd level aggregate (by date) to be the SUM() of aggregates calculated at the first level (by date/tag).
To do that, you can use a window function to compute the AVG() by date/tag before any aggregates are done at all. That makes it possible to SUM() them later. Here is a working example (Oracle 12.1):
-- Create table with test data
create table my_table1 (Date1, tagname, Value) AS (
SELECT TO_DATE('01/01/2021 0:01','DD/MM/YYYY HH24:MI'), 'a', 2 FROM DUAL UNION ALL
SELECT TO_DATE('01/01/2021 0:02','DD/MM/YYYY HH24:MI'), 'a', 4 FROM DUAL UNION ALL
SELECT TO_DATE('01/01/2021 0:01','DD/MM/YYYY HH24:MI'), 'b', 2 FROM DUAL UNION ALL
SELECT TO_DATE('01/01/2021 0:02','DD/MM/YYYY HH24:MI'), 'b', 4 FROM DUAL UNION ALL
SELECT TO_DATE('01/01/2021 0:01','DD/MM/YYYY HH24:MI'), 'c', 2 FROM DUAL UNION ALL
SELECT TO_DATE('01/01/2021 0:02','DD/MM/YYYY HH24:MI'), 'c', 4 FROM DUAL UNION ALL
SELECT TO_DATE('02/01/2021 0:01','DD/MM/YYYY HH24:MI'), 'a', 0 FROM DUAL UNION ALL
SELECT TO_DATE('02/01/2021 0:02','DD/MM/YYYY HH24:MI'), 'a', 0 FROM DUAL UNION ALL
SELECT TO_DATE('02/01/2021 0:01','DD/MM/YYYY HH24:MI'), 'b', 2 FROM DUAL UNION ALL
SELECT TO_DATE('02/01/2021 0:02','DD/MM/YYYY HH24:MI'), 'b', 4 FROM DUAL UNION ALL
SELECT TO_DATE('02/01/2021 0:01','DD/MM/YYYY HH24:MI'), 'c', 2 FROM DUAL UNION ALL
SELECT TO_DATE('02/01/2021 0:02','DD/MM/YYYY HH24:MI'), 'c', 4 FROM DUAL
)
;
-- Compute the averages and the use GROUPING SETS to use those those
-- averages conditionally at multiple levels of aggregation
with date_tag_summary as (
select trunc(date1) date1, tagname, avg(value) avg_value
from my_table1
group by trunc(date1), tagname )
select date1,
case when grouping(tagname)=1 then 'newtag' ELSE tagname END tagname,
case when grouping(tagname)=1 AND COUNT(DECODE(avg_value,0,1,NULL)) > 0 THEN 0
when grouping(tagname)=1 THEN sum(avg_value)
ELSE min(avg_value) END value
from date_tag_summary
group by grouping sets ( (date1, tagname), (date1) )
order by 1,2;
+-----------+---------+-------+
| DATE1 | TAGNAME | VALUE |
+-----------+---------+-------+
| 01-JAN-21 | a | 3 |
| 01-JAN-21 | b | 3 |
| 01-JAN-21 | c | 3 |
| 01-JAN-21 | newtag | 9 |
| 02-JAN-21 | a | 0 |
| 02-JAN-21 | b | 3 |
| 02-JAN-21 | c | 3 |
| 02-JAN-21 | newtag | 0 |
+-----------+---------+-------+
And, to illustrate that the data is not being read twice, here is the execution plan for that query:
-----------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-----------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 6 (100)| |
| 1 | SORT ORDER BY | | 3 | 63 | 6 (50)| 00:00:01 |
| 2 | SORT GROUP BY ROLLUP| | 3 | 63 | 6 (50)| 00:00:01 |
| 3 | VIEW | | 9 | 189 | 4 (25)| 00:00:01 |
| 4 | HASH GROUP BY | | 9 | 117 | 4 (25)| 00:00:01 |
| 5 | TABLE ACCESS FULL| MY_TABLE1 | 12 | 156 | 3 (0)| 00:00:01 |
-----------------------------------------------------------------------------------

Answer 2

One method generates the rows using a cross join and then brings in the existing results:
select d.date1, t.tagname, avg(value) value
from (select distinct to_date(date1, 'dd/MM/yyyy') as date1 from table1
) d cross join
(select 'a' as tagname from dual union all
select 'b' as tagname from dual union all
select 'c' as tagname from dual union all
select 'd' as tagname from dual
) t
table1 t1
on to_date(t1.date1, 'dd/MM/yyyy') = d.date1 and
t1.tagname = t.tagname
group by date1, tagname

Answer 3

You may use grouping sets and then replace avg total for group with sum of averages, calculated by analytic function.
select /*+ gather_plan_statistics */
trunc(date1) as dt
, case grouping_id(tagname)
when 0
then tagname
else 'newtag'
end as tagname
, case grouping_id(tagname)
when 0
then avg(value)
else (
/*Total sum except total avg*/
sum(avg(value)) over(
partition by trunc(date1)
) - avg(value))
* decode(min(avg(value)) over(partition by trunc(date1)), 0, 0, 1)
end as val
from a
group by grouping sets( (trunc(date1), tagname), trunc(date1))
DT | TAGNAME | VAL
:-------- | :------ | --:
01-JAN-21 | a | 3
01-JAN-21 | b | 3
01-JAN-21 | c | 3
01-JAN-21 | newtag | 9
02-JAN-21 | a | 0
02-JAN-21 | b | 3
02-JAN-21 | c | 3
02-JAN-21 | newtag | 0
db<>fiddle here

Answer 4

you can use following query. Of course it is set in SQL
;WITH cte AS
(SELECT convert(date,date1) as date1,tagname,avg(value) value
FROM table1
GROUP BY convert(date,date1),tagname)
select date1,tagname,
case when tagname = 'newtag'
then
case (select cte.value from cte where cte.date1 = result.date1 and cte.tagname = 'a')
when 0 then 0
else (select top 1 sum(c.value) from cte c where convert(date,c.date1,103) = result.date1)
end
else value end
from
(select date1,tagname,value ,ROW_NUMBER() over(partition by date1,tagname order by date1) as seq
from
(
select convert(date,date1) as date1,tagname,avg(value) as value
from table1
group by convert(date,date1),tagname
union all
select convert(date,date1),'newtag', 0
from table1
group by convert(date,date1),tagname
) T
) result
where result.seq = 1
order by convert(date,date1)

Answer 5

first average by day
with avgday as (select avg(value) value, tagname, to_date(date1,'dd/MM/yyyy')
from table1 group by date1, tagname)
transform row's to columns and doing a case to filter and operate.
with query1 as (SELECT * FROM avgday PIVOT ( MAX(value) FOR tagname IN ('a','b','c')))
select date1, case
when query1.a=0
then 0
else a + b + c end value,
'newtag' tagname
from query1
I finally came up with a solution, sure it's not the best answer, but it solves my problem