SQL Grouping by Ranges - sql

Question

I have a data set that has timestamped entries over various sets of groups.
Timestamp -- Group -- Value
---------------------------
1 -- A -- 10
2 -- A -- 20
3 -- B -- 15
4 -- B -- 25
5 -- C -- 5
6 -- A -- 5
7 -- A -- 10
I want to sum these values by the Group field, but parsed as it appears in the data. For example, the above data would result in the following output:
Group -- Sum
A -- 30
B -- 40
C -- 5
A -- 15
I do not want this, which is all I've been able to come up with on my own so far:
Group -- Sum
A -- 45
B -- 40
C -- 5
Using Oracle 11g, this is what I've hobbled togther so far. I know that this is wrong, by I'm hoping I'm at least on the right track with RANK(). In the real data, entries with the same group could be 2 timestamps apart, or 100; there could be one entry in a group, or 100 consecutive. It does not matter, I need them separated.
WITH SUB_Q AS
(SELECT K_ID
, GRP
, VAL
-- GET THE RANK FROM TIMESTAMP TO SEPARATE GROUPS WITH SAME NAME
, RANK() OVER(PARTITION BY K_ID ORDER BY TMSTAMP) AS RNK
FROM MY_TABLE
WHERE K_ID = 123)
SELECT T1.K_ID
, T1.GRP
, SUM(CASE
WHEN T1.GRP = T2.GRP THEN
T1.VAL
ELSE
0
END) AS TOTAL_VALUE
FROM SUB_Q T1 -- MAIN VALUE
INNER JOIN SUB_Q T2 -- TIMSTAMP AFTER
ON T1.K_ID = T2.K_ID
AND T1.RNK = T2.RNK - 1
GROUP BY T1.K_ID
, T1.GRP
Is it possible to group in this way? How would I go about doing this?

Answer 1

I approach this problem by defining a group which is the different of two row_number():
select group, sum(value)
from (select t.*,
(row_number() over (order by timestamp) -
row_number() over (partition by group order by timestamp)
) as grp
from my_table t
) t
group by group, grp
order by min(timestamp);
The difference of two row numbers is constant for adjacent values.

Answer 2

A solution using LAG and windowed analytic functions:
SQL Fiddle
Oracle 11g R2 Schema Setup:
CREATE TABLE TEST ( "Timestamp", "Group", Value ) AS
SELECT 1, 'A', 10 FROM DUAL
UNION ALL SELECT 2, 'A', 20 FROM DUAL
UNION ALL SELECT 3, 'B', 15 FROM DUAL
UNION ALL SELECT 4, 'B', 25 FROM DUAL
UNION ALL SELECT 5, 'C', 5 FROM DUAL
UNION ALL SELECT 6, 'A', 5 FROM DUAL
UNION ALL SELECT 7, 'A', 10 FROM DUAL;
Query 1:
WITH changes AS (
SELECT t.*,
CASE WHEN LAG( "Group" ) OVER ( ORDER BY "Timestamp" ) = "Group" THEN 0 ELSE 1 END AS hasChangedGroup
FROM TEST t
),
groups AS (
SELECT "Group",
VALUE,
SUM( hasChangedGroup ) OVER ( ORDER BY "Timestamp" ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW ) AS grp
FROM changes
)
SELECT "Group",
SUM( VALUE )
FROM Groups
GROUP BY "Group", grp
ORDER BY grp
Results:
| Group | SUM(VALUE) |
|-------|------------|
| A | 30 |
| B | 40 |
| C | 5 |
| A | 15 |

Answer 3

This is typical "star_of_group" problem (see here: https://timurakhmadeev.wordpress.com/2013/07/21/start_of_group/)
In your case, it would be as follows:
with t as (
select 1 timestamp, 'A' grp, 10 value from dual union all
select 2, 'A', 20 from dual union all
select 3, 'B', 15 from dual union all
select 4, 'B', 25 from dual union all
select 5, 'C', 5 from dual union all
select 6, 'A', 5 from dual union all
select 7, 'A', 10 from dual
)
select min(timestamp), grp, sum(value) sum_value
from (
select t.*
, sum(start_of_group) over (order by timestamp) grp_id
from (
select t.*
, case when grp = lag(grp) over (order by timestamp) then 0 else 1 end
start_of_group
from t
) t
)
group by grp_id, grp
order by min(timestamp)
;