I'm trying to create a query that contains a running count of ISO weeks when the account has no entry in a [Volume] table. The query is to return only a sample of the accounts, so I've created a couple of CTEs to limit the number of records and join the volume table so that weeks with no volume show up in the results.
To illustrate the result I'm looking for:
ISOWk | SurrID | Weekly Volume | No vol Count
201601 | 001 | 0 | 1
201601 | 002 | 5 | 0
201602 | 001 | 0 | 2
201602 | 002 | 0 | 1
201603 | 001 | 125| 0
201603 | 002 | 75 | 0
201604 | 001 | 0 | 1
201604 | 002 | 75 | 0
As you can see, account with SurrID 001 has no volume for weeks 201601 and 201602 so has [No vol Count] of 2 in week 201602. In week 201603, there is volume so the counter resets to 0 and increases to 1 week 201604.
From the research I've done I've managed to get a consecutive count running using ROW_NUMBER with a window function, but it doesn't reset if there is volume (as in week 201603 in my example). What I can't figure out is how to count consecutive zero values and reset when needed.
I've included my full query below so that you get the full picture (please point out any particularly bad practices here - I'm still finding my way!). It all works as intended until I include the third CTE "NDs". It then takes 45 mins to return just over 2000 rows and returns an non-resetting count for rows showing no weekly volume.
WITH Surrs AS
(
SELECT SurrID, OracleStartDate AS OSD
FROM (
SELECT ca.SurrID, OracleStartDate, ROW_NUMBER() OVER(ORDER BY OracleStartDate) as rn
FROM tblCustomerAccounts ca
JOIN tblAccountUpdates au
ON ca.SurrID = au.SurrID
WHERE CustomerType_ID IN (1,2,3,4,5,6,7,12)
AND au.ISOWk = 201641
) a
WHERE rn % 1000 = 0
),
Updates AS
(
SELECT au.ISOWk, s.SurrID, (CASE WHEN AccStatus_ID = 1 THEN 'A' ELSE 'I' END) AS AccStatus, (CASE WHEN dbo.udf_ConvertDateToISOWeek(OSD) <= BBC THEN 'B' ELSE 'F' END) AS Book
FROM Surrs s
JOIN (
SELECT ISOWk,
(SELECT BBCutOff FROM dbo.udf_CutOffWeeks(ISOWk)) AS BBC,
(SELECT FYStart FROM dbo.udf_CutOffWeeks(ISOWk)) AS FYS,
(SELECT FYEnd FROM dbo.udf_CutOffWeeks(ISOWk)) AS FYE,
(SELECT BYStart FROM dbo.udf_CutOffWeeks(ISOWk)) AS BYS,
(SELECT BYEnd FROM dbo.udf_CutOffWeeks(ISOWk)) AS BYE,
SurrID,
AccStatus_ID
FROM tblAccountUpdates
) au
ON au.SurrID = s.SurrID
),
NDs AS
(
SELECT u.ISOWk, u.SurrID, ROW_NUMBER() OVER (PARTITION BY u.SurrID ORDER BY u.ISOWk) AS NDCount
FROM Updates u
LEFT JOIN tblTotalVolumes tv
ON u.SurrID = tv.SurrID
AND u.ISOWk = tv.ISOWk
WHERE tv.Volume IS NULL
AND u.ISOWk >= 201601
)
SELECT tw.ISOWk,
tw.SurrID,
(CASE WHEN Volume IS NULL THEN 0 ELSE Volume END) AS [Weekly Volume],
tw.Book,
tw.AccStatus,
(CASE WHEN tw.AccStatus = 'I' AND lw.AccStatus = 'A' THEN 'Y' ELSE '' END) AS [Stopped this week],
(CASE WHEN tw.AccStatus = 'A' AND lw.AccStatus = 'I' THEN 'Y' ELSE '' END) AS [Restarted this week],
(CASE WHEN NDCount IS NULL THEN 0 ELSE NDCount END) AS [Consecutive ND Weeks]
FROM Updates tw
JOIN Updates lw
ON lw.ISOWk = dbo.udf_ConvertDateToISOWeek(DATEADD("ww",-1,dbo.udf_ConvertISOWkToDate(tw.ISOWk)))
AND tw.SurrID = lw.SurrID
LEFT JOIN tblTotalVolumes tv
ON tw.SurrID = tv.SurrID
AND tw.ISOWk = tv.ISOWk
LEFT JOIN NDs
ON tw.SurrID = nds.SurrID
AND tw.ISOWk = nds.ISOWk
ORDER BY tw.ISOWk
To reiterate what I need: the column [Consecutive ND Weeks] should count consecutive weeks where [Weekly Volume] is 0. Help will be much appreciated.
Thanks
UPDATE:
I've tried to implement #Gordon Linoff's post but my counter does not reset to 0 when there is a value for [Weekly Volume]. Here's my amended query:
SELECT t.*, (CASE WHEN [Weekly Volume] = 0 THEN ROW_NUMBER() OVER (PARTITION BY t.SurrID, grp ORDER BY ISOWk) ELSE 0 END) AS [ND Count]
FROM (
SELECT tw.ISOWk,
s.SurrID,
tw.AccStatus,
(CASE WHEN tv.Volume IS NULL THEN 0 ELSE tv.Volume END) AS [Weekly Volume],
(CASE WHEN dbo.udf_ConvertDateToISOWeek(OSD) <= BBC THEN 'B' ELSE 'F' END) AS Book,
(CASE WHEN tw.AccStatus = 'I' AND lw.AccStatus = 'A' THEN 'Y' ELSE '' END) AS [Stopped this week],
(CASE WHEN tw.AccStatus = 'A' AND lw.AccStatus = 'I' THEN 'Y' ELSE '' END) AS [Restarted this week],
SUM(CASE WHEN tv.volume > 0 THEN 1 ELSE 0 END) OVER(PARTITION BY tv.SurrID ORDER BY tv.ISOWk) AS grp
FROM (
SELECT SurrID, OracleStartDate AS OSD
FROM (
SELECT ca.SurrID, OracleStartDate, ROW_NUMBER() OVER(ORDER BY OracleStartDate) as rn
FROM tblCustomerAccounts ca
JOIN tblAccountUpdates au
ON ca.SurrID = au.SurrID
WHERE CustomerType_ID IN (1,2,3,4,5,6,7,12)
AND au.ISOWk = 201641
) a
WHERE rn % 1000 = 0
) s
JOIN (
SELECT ISOWk,
(SELECT BBCutOff FROM dbo.udf_CutOffWeeks(ISOWk)) AS BBC,
(SELECT FYStart FROM dbo.udf_CutOffWeeks(ISOWk)) AS FYS,
(SELECT FYEnd FROM dbo.udf_CutOffWeeks(ISOWk)) AS FYE,
(SELECT BYStart FROM dbo.udf_CutOffWeeks(ISOWk)) AS BYS,
(SELECT BYEnd FROM dbo.udf_CutOffWeeks(ISOWk)) AS BYE,
SurrID,
(CASE WHEN AccStatus_ID = 1 THEN 'A' ELSE 'I' END) AS AccStatus
FROM tblAccountUpdates
) tw
ON tw.SurrID = s.SurrID
JOIN (
SELECT ISOWk,
SurrID,
(CASE WHEN AccStatus_ID = 1 THEN 'A' ELSE 'I' END) AS AccStatus
FROM tblAccountUpdates
) lw
ON tw.SurrID = lw.SurrID
AND dbo.udf_ConvertDateToISOWeek(DATEADD("ww",-1,dbo.udf_ConvertISOWkToDate(tw.ISOWk))) = lw.ISOWk
LEFT JOIN tblTotalVolumes tv
ON tw.ISOWk = tv.ISOWk
AND tw.SurrID = tv.SurrID
) t
ORDER BY ISOWk
UPDATE:
I've now modified my query to reflect Vladimir's solution (once again, this is the full query):
SELECT ISOWk,
SurrID,
AccStatus,
[Weekly Volume],
Book,
[Stopped this week],
[Restarted this week],
RN1,
RN2,
grp,
rn3,
(CASE WHEN [Weekly Volume] = 0 THEN rn3 ELSE 0 END) AS [ND Count]
FROM (
SELECT t.ISOWk,
t.SurrID,
t.AccStatus,
t.[Weekly Volume],
t.Book,
t.[Stopped this week],
t.[Restarted this week],
rn1,
rn2,
rn1 - rn2 AS grp,
ROW_NUMBER() OVER(PARTITION BY t.SurrID, rn1-rn2 ORDER BY ISOWk) AS rn3
FROM (
SELECT tw.ISOWk,
s.SurrID,
tw.AccStatus,
(CASE WHEN tv.Volume IS NULL THEN 0 ELSE tv.Volume END) AS [Weekly Volume],
(CASE WHEN dbo.udf_ConvertDateToISOWeek(OSD) <= BBC THEN 'B' ELSE 'F' END) AS Book,
(CASE WHEN tw.AccStatus = 'I' AND lw.AccStatus = 'A' THEN 'Y' ELSE '' END) AS [Stopped this week],
(CASE WHEN tw.AccStatus = 'A' AND lw.AccStatus = 'I' THEN 'Y' ELSE '' END) AS [Restarted this week],
ROW_NUMBER() OVER(PARTITION BY tw.SurrID ORDER BY tw.ISOWk) AS rn1,
ROW_NUMBER() OVER(PARTITION BY tw.SurrID, tv.Volume ORDER BY tw.ISOWk) AS rn2
FROM (
SELECT SurrID, OracleStartDate AS OSD
FROM (
SELECT ca.SurrID, OracleStartDate, ROW_NUMBER() OVER(ORDER BY OracleStartDate) as rn
FROM tblCustomerAccounts ca
JOIN tblAccountUpdates au
ON ca.SurrID = au.SurrID
WHERE CustomerType_ID IN (1,2,3,4,5,6,7,12)
AND au.ISOWk = 201641
) a
WHERE rn % 2000 = 0
) s
JOIN (
SELECT ISOWk,
(SELECT BBCutOff FROM dbo.udf_CutOffWeeks(ISOWk)) AS BBC,
(SELECT FYStart FROM dbo.udf_CutOffWeeks(ISOWk)) AS FYS,
(SELECT FYEnd FROM dbo.udf_CutOffWeeks(ISOWk)) AS FYE,
(SELECT BYStart FROM dbo.udf_CutOffWeeks(ISOWk)) AS BYS,
(SELECT BYEnd FROM dbo.udf_CutOffWeeks(ISOWk)) AS BYE,
SurrID,
(CASE WHEN AccStatus_ID = 1 THEN 'A' ELSE 'I' END) AS AccStatus
FROM tblAccountUpdates
) tw
ON tw.SurrID = s.SurrID
JOIN (
SELECT ISOWk,
SurrID,
(CASE WHEN AccStatus_ID = 1 THEN 'A' ELSE 'I' END) AS AccStatus
FROM tblAccountUpdates
) lw
ON tw.SurrID = lw.SurrID
AND dbo.udf_ConvertDateToISOWeek(DATEADD("ww",-1,dbo.udf_ConvertISOWkToDate(tw.ISOWk))) = lw.ISOWk
LEFT JOIN tblTotalVolumes tv
ON tw.ISOWk = tv.ISOWk
AND tw.SurrID = tv.SurrID
) t
) x
ORDER BY ISOWk
Here's a sample of the unexpected (in that the counter doesn't reset to 0 when [Weekly Volume] has a value greater than 0, identified with the asterisks) results. The following all have the same ID, so I've removed the ID column.
ISOWk | Weekly Volume | rn1 | rn2 | grp | rn3 | ND Count |
201620 | 0 | 1 | 1 | 0 | 1 | 1 |
201621 | 0 | 2 | 2 | 0 | 2 | 2 |
201622 | 0 | 3 | 3 | 0 | 3 | 3 |
201623 | 0 | 4 | 4 | 0 | 4 | 4 |
201624 | 0 | 5 | 5 | 0 | 5 | 5 |
201625 | 53 | 6 | 1 | 5 | 1 | 0 |
201626 | 49 | 7 | 1 | 6 | 1 | 0 |
201627 | 98 | 8 | 1 | 7 | 1 | 0 |
201628 | 54 | 9 | 1 | 8 | 1 | 0 |
201629 | 53 | 10 | 2 | 8 | 2 | 0 |
201630 | 103 | 11 | 1 | 10 | 1 | 0 |
201631 | 59 | 12 | 1 | 11 | 1 | 0 |
201632 | 35 | 13 | 1 | 12 | 1 | 0 |
201633 | 0 | 14 | 6 | 8 | 3 | 3 |**
201634 | 0 | 15 | 7 | 8 | 4 | 4 |**
201635 | 0 | 16 | 8 | 8 | 5 | 5 |**
201636 | 0 | 17 | 9 | 8 | 6 | 6 |**
201637 | 87 | 18 | 1 | 17 | 1 | 0 |
201638 | 136 | 19 | 1 | 18 | 1 | 0 |
201639 | 56 | 20 | 1 | 19 | 1 | 0 |
201640 | 70 | 21 | 1 | 20 | 0 | 0 |
201641 | 77 | 22 | 1 | 21 | 1 | 0 |
There are other instances of this issue in my dataset.
It looks like a gaps-and-islands problem.
Sample data
DECLARE #T TABLE(ISOWk int, SurrID char(3), WeeklyVolume int);
INSERT INTO #T(ISOWk, SurrID, WeeklyVolume) VALUES
(201601, '001', 0),
(201601, '002', 5),
(201602, '001', 0),
(201602, '002', 0),
(201603, '001',125),
(201603, '002', 75),
(201604, '001', 0),
(201604, '002', 75),
(201620, '003', 0),
(201621, '003', 0),
(201622, '003', 0),
(201623, '003', 0),
(201624, '003', 0),
(201625, '003', 53),
(201626, '003', 49),
(201627, '003', 98),
(201628, '003', 54),
(201629, '003', 53),
(201630, '003',103),
(201631, '003', 59),
(201632, '003', 35),
(201633, '003', 0),
(201634, '003', 0),
(201635, '003', 0),
(201636, '003', 0),
(201637, '003', 87),
(201638, '003',136),
(201639, '003', 56),
(201640, '003', 70),
(201641, '003', 77),
(201601, '004', 0),
(201602, '004', 6),
(201603, '004', 0),
(201604, '004', 0);
I added your extended sample with SurrID=003 and added mine sample with SurrID=004.
Query
WITH
CTE
AS
(
SELECT
ISOWk
,SurrID
,WeeklyVolume
,ROW_NUMBER() OVER (PARTITION BY SurrID ORDER BY ISOWk) AS rn1
,ROW_NUMBER() OVER (PARTITION BY SurrID,WeeklyVolume ORDER BY ISOWk) AS rn2
FROM #T
)
,CTE2
AS
(
SELECT
ISOWk
,SurrID
,WeeklyVolume
,rn1
,rn2
,rn1-rn2 AS grp
,ROW_NUMBER() OVER (PARTITION BY SurrID,WeeklyVolume,rn1-rn2 ORDER BY ISOWk) AS rn3
FROM CTE
)
SELECT
ISOWk
,SurrID
,WeeklyVolume
,rn1
,rn2
,grp
,rn3
,CASE WHEN WeeklyVolume = 0 THEN rn3 ELSE 0 END AS NoVolumeCount
FROM CTE2
ORDER BY SurrID, ISOWk;
Result
+--------+--------+--------------+-----+-----+-----+-----+---------------+
| ISOWk | SurrID | WeeklyVolume | rn1 | rn2 | grp | rn3 | NoVolumeCount |
+--------+--------+--------------+-----+-----+-----+-----+---------------+
| 201601 | 001 | 0 | 1 | 1 | 0 | 1 | 1 |
| 201602 | 001 | 0 | 2 | 2 | 0 | 2 | 2 |
| 201603 | 001 | 125 | 3 | 1 | 2 | 1 | 0 |
| 201604 | 001 | 0 | 4 | 3 | 1 | 1 | 1 |
| 201601 | 002 | 5 | 1 | 1 | 0 | 1 | 0 |
| 201602 | 002 | 0 | 2 | 1 | 1 | 1 | 1 |
| 201603 | 002 | 75 | 3 | 1 | 2 | 1 | 0 |
| 201604 | 002 | 75 | 4 | 2 | 2 | 2 | 0 |
| 201620 | 003 | 0 | 1 | 1 | 0 | 1 | 1 |
| 201621 | 003 | 0 | 2 | 2 | 0 | 2 | 2 |
| 201622 | 003 | 0 | 3 | 3 | 0 | 3 | 3 |
| 201623 | 003 | 0 | 4 | 4 | 0 | 4 | 4 |
| 201624 | 003 | 0 | 5 | 5 | 0 | 5 | 5 |
| 201625 | 003 | 53 | 6 | 1 | 5 | 1 | 0 |
| 201626 | 003 | 49 | 7 | 1 | 6 | 1 | 0 |
| 201627 | 003 | 98 | 8 | 1 | 7 | 1 | 0 |
| 201628 | 003 | 54 | 9 | 1 | 8 | 1 | 0 |
| 201629 | 003 | 53 | 10 | 2 | 8 | 1 | 0 |
| 201630 | 003 | 103 | 11 | 1 | 10 | 1 | 0 |
| 201631 | 003 | 59 | 12 | 1 | 11 | 1 | 0 |
| 201632 | 003 | 35 | 13 | 1 | 12 | 1 | 0 |
| 201633 | 003 | 0 | 14 | 6 | 8 | 1 | 1 |
| 201634 | 003 | 0 | 15 | 7 | 8 | 2 | 2 |
| 201635 | 003 | 0 | 16 | 8 | 8 | 3 | 3 |
| 201636 | 003 | 0 | 17 | 9 | 8 | 4 | 4 |
| 201637 | 003 | 87 | 18 | 1 | 17 | 1 | 0 |
| 201638 | 003 | 136 | 19 | 1 | 18 | 1 | 0 |
| 201639 | 003 | 56 | 20 | 1 | 19 | 1 | 0 |
| 201640 | 003 | 70 | 21 | 1 | 20 | 1 | 0 |
| 201641 | 003 | 77 | 22 | 1 | 21 | 1 | 0 |
| 201601 | 004 | 0 | 1 | 1 | 0 | 1 | 1 |
| 201602 | 004 | 6 | 2 | 1 | 1 | 1 | 0 |
| 201603 | 004 | 0 | 3 | 2 | 1 | 1 | 1 |
| 201604 | 004 | 0 | 4 | 3 | 1 | 2 | 2 |
+--------+--------+--------------+-----+-----+-----+-----+---------------+
I included intermediate columns in the result, so you can understand how it works.
Standard approach for gaps-and-islands is two sequences of ROW_NUMBER - one is a plain sequence (rn1), second is partitioned by WeeklyVolume (rn2).
The difference between rn1 and rn2 gives the ID of the group (island) (grp). Calculate another sequence of row numbers partitioned by WeeklyVolume and this group (rn3) and use it only when WeeklyVolume is zero.
Obviously, everything above is partitioned by SurrID at first.
In the first variant of the answer I forgot to include WeeklyVolume in the last partitioning for rn3.
Here is an approach:
Count the number of preceding non-0 values for SurrId. This number identifies groups.
Do a row_number() within each group.
Only consider the row_number() when the value is 0.
This results in:
select t.*,
(case when weeklyvolume = 0
then row_number() over (partition by SurrId, grp order by ISOwk)
else 0
end) as NoVolCount
from (select t.*,
sum(case when weeklyvolume > 0 then 1 else 0 end) over (partition by SurrId order by ISOwk) as grp
from t
) t;
Related
I have a table of Account Transactions that includes ID, Amount, Date. Basically, I want to create a resulting table that looks at the table and returns what the SUM was for the Account over three different Ending Date Ranges. Then I want to Flag (Combined_Flag) each Account ID, 1 if any of the SUMs for that ID are non-zero, and a 0 if all of the SUMs are 0.
Date Range 1) Min Date to End of Last Month (-1 Month)
Date Range 2) Min Date to End of 2 Months ago (-2 Months)
Date Range 3) Min Date to End of Last Month, Last Year (-13 Months)
The Resulting table should be: ID, SUM_R1, SUM_R2, SUM_R3, Flag_R1, Flag_R2, Flag_R3, Combined_Flag
Example Data
| ID | Amount | Date |
| -------- | -------------- |-------------- |
| 1 | 20 | 09/01/19 |
| 2 | 40 | 09/01/19 |
| 3 | 0 | 09/01/19 |
| 4 | 0 | 09/01/19 |
| 1 | 10 | 10/01/19 |
| 2 | 0 | 10/01/19 |
| 3 | 0 | 10/01/19 |
| 4 | 0 | 10/01/19 |
| 1 | 15 | 11/01/19 |
| 2 | 40 | 11/01/19 |
| 3 | 0 | 11/01/19 |
| 4 | 0 | 11/01/19 |
| 1 | 20 | 09/01/20 |
| 2 | 40 | 09/01/20 |
| 3 | 0 | 09/01/20 |
| 4 | 50 | 09/01/20 |
| 1 | 10 | 10/01/20 |
| 2 | 0 | 10/01/20 |
| 3 | 0 | 10/01/20 |
| 4 | 65 | 10/01/20 |
| 1 | 15 | 11/01/20 |
| 2 | 40 | 11/01/20 |
| 3 | 0 | 11/01/20 |
| 4 | 0 | 11/01/20 |
Expected Result Table (Using Date of 12/21/2020)
| ID | SUM_R1 | SUM_R2 | SUM_R3 | Flag_R1 | Flag_R2 | Flag_R3 | Combined_Flag |
| -------- | -------- | -------- | -------- | --------- | --------- | --------- | --------------- |
| 1 | 90 | 75 | 45 | 1 | 1 | 1 | 1 |
| 2 | 160 | 120 | 80 | 1 | 1 | 1 | 1 |
| 3 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 4 | 115 | 115 | 0 | 1 | 1 | 0 | 1 |
The difficulty I'm having is in joining the table basically to itself 2 times. I'm getting results all over the place and not really sure exactly what's going on.
Is this what you want?
select id,
sum(case when date < datefromparts(year(v.dte), month(v.dte), 1)
then amount else 0
end) as sum_r1,
sum(case when date < dateadd(month, -1, datefromparts(year(v.dte), month(v.dte), 1))
then amount else 0
end) as sum_r2,
sum(case when date < dateadd(month, -13, datefromparts(year(v.dte), month(v.dte), 1))
then amount else 0
end) as sum_r3,
max(case when amount > 0 and date < datefromparts(year(v.dte), month(v.dte), 1)
then 1 else 0
end) as flag_r1,
max(case when amount > 0 and date < dateadd(month, -1, datefromparts(year(v.dte), month(v.dte), 1))
then 1 else 0
end) as flag_r2,
max(case when amount > 0 and date < dateadd(month, -13, datefromparts(year(v.dte), month(v.dte), 1))
then 1 else 0
end) as flag_r3
from t cross join
(values (convert(date, '2020-12-21'))
) v(dte)
group by id;
The flag columns assume that the amounts are never negative (which is consistent with the data in your question.
EDIT:
The shorthand in the comment for creating the flag looks like:
abs(sign(sum(case when amount > 0 and date < datefromparts(year(v.dte), month(v.dte), 1)
then amount else 0
end))) as flag_r1,
Here is a db<>fiddle.
How can I get second largest column value and its name?
My current query gives it mostly correct but in cases where largest and second largest values are same I am getting wrong values.
select item_code, A, B, C,
greatest(A, B, C) as largest1,
greatest(case when largest1 = A then 0 else A end,
case when largest1 = B then 0 else B end,
case when largest1 = C then 0 else C end) as largest2,
(case largest1 when A then 'A'
when B then 'B'
when C then 'C' end) as largest1_column_name,
(case largest2 when A then 'A'
when B then 'B'
when C then 'C' else 'None' end) as largest2_column_name
from table1
Below is the sample table:
+-----------+----+----+----+
| item_code | A | B | C |
+-----------+----+----+----+
| p1 | 20 | 30 | 40 |
| p2 | 50 | 30 | 10 |
| p3 | 30 | 50 | 10 |
| p4 | 30 | 30 | 30 |
| p5 | 50 | 50 | 10 |
| p6 | 0 | 0 | 0 |
+-----------+----+----+----+
Below is expected output:
+-----------+----+----+----+----------+----------+----------------------+----------------------+
| item_code | A | B | C | largest1 | largest2 | largest1_column_name | largest2_column_name |
+-----------+----+----+----+----------+----------+----------------------+----------------------+
| p1 | 20 | 30 | 40 | 40 | 30 | C | B |
| p2 | 50 | 30 | 10 | 50 | 30 | A | B |
| p3 | 30 | 50 | 10 | 50 | 30 | B | A |
| p4 | 30 | 30 | 30 | 30 | 30 | A | B |
| p5 | 50 | 50 | 10 | 50 | 50 | A | B |
| p6 | 0 | 0 | 0 | 0 | 0 | A | B |
+-----------+----+----+----+----------+----------+----------------------+----------------------+
This is the output I am getting from my query (I have marked wrong as comment):
+-----------+----+----+----+----------+-------------+----------------------+----------------------+
| item_code | A | B | C | largest1 | largest2 | largest1_column_name | largest2_column_name |
+-----------+----+----+----+----------+-------------+----------------------+----------------------+
| p1 | 20 | 30 | 40 | 40 | 30 | C | B |
| p2 | 50 | 30 | 10 | 50 | 30 | A | B |
| p3 | 30 | 50 | 10 | 50 | 30 | B | A |
| p4 | 30 | 30 | 30 | 30 | 0/*wrong*/ | A | NULL/*wrong*/ |
| p5 | 50 | 50 | 10 | 50 | 10/*wrong*/ | A | C/*wrong*/ |
| p6 | 0 | 0 | 0 | 0 | 0/*wrong*/ | A | A/*wrong*/ |
+-----------+----+----+----+----------+-------------+----------------------+----------------------+
I tried a slight variation of this (listagg instead of string_agg) in Snowflake and it seemed to be getting the expected result
with cte (item_code, abc, id) as
(select item_code, a, 'a' from table1 union all
select item_code, b, 'b' from table1 union all
select item_code, c, 'c' from table1)
select item_code,
max(case when id='a' then abc end) a,
max(case when id='b' then abc end) b,
max(case when id='c' then abc end) c,
split_part(string_agg(abc::varchar,',' order by abc desc),',',1) largest1,
split_part(string_agg(abc::varchar,',' order by abc desc),',',2) largest2,
split_part(string_agg(id,',' order by abc desc),',',1) largest1_col,
split_part(string_agg(id,',' order by abc desc),',',2) largest2_col
from cte
group by item_code;
This might be simpler achieved by unpivoting the rows, ranking the values, and then using conditional aggregation. In Postgres, you could phrase this as:
select t.*, x.*
from table1 t1
cross join lateral (
select
min(val) filter(where rn = 1) largest1,
min(val) filter(where rn = 2) largest2,
min(col) filter(where rn = 1) largest1_column_name,
min(col) filter(where rn = 2) largest2_column_name
from (
select x.*, dense_rank() over(order by val desc) rn
from (values ('a', a), ('b', b), ('c', c)) as x(col, val)
) x
) x
What I'm trying to achieve is a cumulative sum on a non-negative column where it decreases by 1 on every row, however the result must also be non-negative.
For example, for the following table, summing over the "VALUE" column ordered by the "ID" column:
| ID | VALUE |
-----------------
| 1 | 0 |
| 2 | 0 |
| 3 | 2 |
| 4 | 0 |
| 5 | 0 |
| 6 | 3 |
| 7 | 0 |
| 8 | 2 |
| 9 | 0 |
| 10 | 0 |
| 11 | 0 |
| 12 | 0 |
I expect:
| ID | VALUE | SUM |
-------------------------
| 1 | 0 | 0 |
| 2 | 0 | 0 |
| 3 | 2 | 2 |
| 4 | 0 | 1 |
| 5 | 0 | 0 |
| 6 | 3 | 3 |
| 7 | 0 | 2 |
| 8 | 2 | 3 |
| 9 | 0 | 2 |
| 10 | 0 | 1 |
| 11 | 0 | 0 |
| 12 | 0 | 0 |
Your question is not very well described. My best interpretation is that you want to count down from positive value, starting over when you hit the next one.
You can define the groups with a cumulative sum of the non-zero values. Then use a cumulative sum on the groups:
select t.*,
(case when max(value) over (partition by grp) < row_number() over (partition by grp order by id) - 1
then 0
else (max(value) over (partition by grp) -
(row_number() over (partition by grp order by id) - 1)
)
end) as my_value
from (select t.*,
sum(case when value <> 0 then 1 else 0 end) over (order by id) as grp
from t
) t
Here is a db<>fiddle.
EDIT:
It strikes me that you might want to keep all the "positive" values and count down -- remembering if they don't go down to zero. Alas, I think the simplest method is a recursive CTE in this case:
with tn as (
select t.id, t.value, row_number() over (order by id) as seqnum
from t
),
cte as (
select tn.id, tn.value, tn.seqnum, tn.value as s
from tn
where id = 1
union all
select tn.id, tn.value, tn.seqnum,
(case when cte.s = 0
then tn.value
when tn.value = 0 and cte.s > 0
then cte.s - 1
-- when tn.value > 0 and cte.value > 0
else tn.value + cte.s - 1
end)
from cte join
tn
on tn.seqnum = cte.seqnum + 1
)
select *
from cte;
The db<>fiddle has both solutions.
I have a table with following records:
=======================
| ID | device_num |
=======================
| 1 | 11 |
| 2 | 11 |
| 3 | NULL |
| 4 | 11 |
| 5 | 11 |
| 6 | NULL |
| 7 | 11 |
| 8 | 12 |
| 9 | 12 |
| 10 | 13 |
| 11 | NULL |
| 12 | 13 |
| 13 | 13 |
| 14 | 13 |
| 15 | 14 |
| 16 | 14 |
=======================
I want to assign a rank to each device number based on following cases:
1- Rank should be started with 1.
2- Each record should be compared with its previous record if both has the same device number then assign a same rank to records.
3- Do not assign rank to NULL records. In case if we get same device number after the Null record then rank should be increased by one.
4- If device number not matches with previous record then increase rank by 1.
Desired output:
===============================
| ID | RANK | device_num |
===============================
| 1 | 1 | 11 |
| 2 | 1 | 11 |
| 3 | | NULL |
| 4 | 2 | 11 |
| 5 | 2 | 11 |
| 6 | | NULL |
| 7 | 3 | 11 |
| 8 | 4 | 12 |
| 9 | 4 | 12 |
| 10 | 5 | 13 |
| 11 | | NULL |
| 12 | 6 | 13 |
| 13 | 6 | 13 |
| 14 | 6 | 13 |
| 15 | 7 | 14 |
| 16 | 7 | 14 |
===============================
I have tried with PostgreSQL Rank functions dense,rank etc but not able to assign rank in this order.
Try this:
SELECT id, device_num,
CASE
WHEN device_num IS NOT NULL
THEN DENSE_RANK() OVER (ORDER BY CASE
WHEN device_num IS NOT NULL
THEN min_id
END)
END AS RANK
FROM (
SELECT id, device_num,
MIN(id) OVER (PARTITION BY device_num, grp) AS min_id
FROM (
SELECT id, device_num,
ROW_NUMBER() OVER (ORDER BY id) -
ROW_NUMBER() OVER (PARTITION BY CASE
WHEN device_num IS NULL THEN -1
ELSE device_num
END
ORDER BY id) AS grp
FROM mytable) AS t) AS s
ORDER BY id
I've made the assumptions that:
there is an auto increment id field in your table that specifies row order.
device_num field is of integer type and -1 is not a valid value for the field.
Demo here
I write a query like this:
;WITH t1 AS (
SELECT *
, ROW_NUMBER() OVER (ORDER BY (SELECT null)) AS rn
FROM yourTable
) -- To add a row-number to yourTable
, t2 AS (
SELECT *
, CASE
WHEN device_num IS NULL THEN 0
WHEN ISNULL(LAG(device_num) OVER (ORDER BY rn), -1) <> ISNULL(device_num, -1) THEN 1
ELSE 0
END AS willChange
FROM t1
)
SELECT device_num
, CASE
WHEN device_num IS NULL THEN null
ELSE SUM(willChange) OVER (ORDER BY rn)
END [RANK]
FROM t2;
EDIT : By using ID: it can be like this:
;WITH t AS (
SELECT *
, CASE
WHEN device_num IS NULL THEN 0
WHEN ISNULL(LAG(device_num) OVER (ORDER BY ID), -1) <> ISNULL(device_num, -1) THEN 1
ELSE 0
END AS willChange
FROM yourTable
)
SELECT device_num
, CASE
WHEN device_num IS NULL THEN null
ELSE SUM(willChange) OVER (ORDER BY ID)
END [RANK]
FROM t;
I need help with a SQL that will convert this table:
===================
| Id | FK | Status|
===================
| 1 | A | 100 |
| 2 | A | 101 |
| 3 | B | 100 |
| 4 | B | 101 |
| 5 | C | 100 |
| 6 | C | 101 |
| 7 | A | 102 |
| 8 | A | 102 |
| 9 | B | 102 |
| 10 | B | 102 |
===================
to this:
==========================================
| FK | Count 100 | Count 101 | Count 102 |
==========================================
| A | 1 | 1 | 2 |
| B | 1 | 1 | 2 |
| C | 1 | 1 | 0 |
==========================================
I can so simple counts, etc., but am struggling trying to pivot the table with the information derived. Any help is appreciated.
Use:
SELECT t.fk,
SUM(CASE WHEN t.status = 100 THEN 1 ELSE 0 END) AS count_100,
SUM(CASE WHEN t.status = 101 THEN 1 ELSE 0 END) AS count_101,
SUM(CASE WHEN t.status = 102 THEN 1 ELSE 0 END) AS count_102
FROM TABLE t
GROUP BY t.fk
use:
select * from
(select fk,fk as fk1,statusFK from #t
) as t
pivot
(COUNT(fk1) for statusFK IN ([100],[101],[102])
) AS pt
Just adding a shortcut to #OMG's answer.
You can eliminate CASE statement:
SELECT t.fk,
SUM(t.status = 100) AS count_100,
SUM(t.status = 101) AS count_101,
SUM(t.status = 102) AS count_102
FROM TABLE t
GROUP BY t.fk