How make Excel arbitrary pivoting in MS SQL 2017 (cross join + loops) - sql

Question

Would you help me please, to solve the task below in SQL (MS SQL Server 2017). It is simple in Excel, but seems very complicated in SQL.
There is a table with clients and their activities split by days:
client 1may 2may 3may 4may 5may other days
client1 0 0 0 0 0 ...
client2 0 0 0 0 0 ...
client3 0 0 0 0 0 ...
client4 1 1 1 1 1 ...
client5 1 1 1 0 0 ...
It is necessary to create the same table (the same quantity of rows and columns), but turn the values into new one according to the rule:
Current day value =
A) If all everyday values during a week before the day, including the current one = 1, then 1
B) If all everyday values during a week before the day, including the current one = 0, then 0
C) If the values are different, then we leave the status of the previous day (if the status of the previous day is not known, for example, the Client is new, then 0)
In Excel, I do this using the formula: = IF (AND (AF2 = AE2; AE2 = AD2; AD2 = AC2; AC2 = AB2; AB2 = AA2; AA2 = Z2); current_day_value; IF (previous_day_value = ""; 0; previous_day_value )).
The example with excel file is attached.
Thank you very much.

Answer 1

First thing, it's NEVER a good idea to have dates as columns.
So step #1 transpose your columns to rows. In other world to build a table with three columns
```
client date Value
client1 May1 0
client1 May2 0
client1 May3 0
.... ... ..
client4 May1 1
client4 May2 1
client4 May3 1
.... ... ..
```
step #2 perform all the calculations you need by using the date field.

Answer 2

Basically you put always the status of the previous day, in any case (except null).
So, i would do something like this (oracle syntax, working in sql server too), supposing the first columns is 1may
Insert into newTable (client, 1may,2may,....) select (client, 0, coalesce(1may,0), coalesce (2may,0), .... from oldTable;
Anyway me too i believe is not a good practice to put the days as columns of a relational table.

Answer 3

You're going to struggle with this because most brands of SQL don't allow "arbitrary pivoting", that is, you need to specify the columns you want to be displayed on a pivot - Whereas Excel will just do this for you. SQL can do this but it required dynamic SQL which can get pretty complicated and annoying pretty fast.
I would suggest you use sql just to construct the data, and then excel or SSRS (As you're in TSQL) to actually do the visualization.
Anyway. I think this does what you want:
WITH Data AS (
SELECT * FROM (VALUES
('Client 1',CONVERT(DATE, '2020-05-04'),1)
, ('Client 1',CONVERT(DATE, '2020-05-05'),1)
, ('Client 1',CONVERT(DATE, '2020-05-06'),1)
, ('Client 1',CONVERT(DATE, '2020-05-07'),0)
, ('Client 1',CONVERT(DATE, '2020-05-08'),0)
, ('Client 1',CONVERT(DATE, '2020-05-09'),0)
, ('Client 1',CONVERT(DATE, '2020-05-10'),1)
, ('Client 1',CONVERT(DATE, '2020-05-11'),1)
, ('Client 1',CONVERT(DATE, '2020-05-12'),1)
, ('Client 2',CONVERT(DATE, '2020-05-04'),1)
, ('Client 2',CONVERT(DATE, '2020-05-05'),0)
, ('Client 2',CONVERT(DATE, '2020-05-06'),0)
, ('Client 2',CONVERT(DATE, '2020-05-07'),1)
, ('Client 2',CONVERT(DATE, '2020-05-08'),0)
, ('Client 2',CONVERT(DATE, '2020-05-09'),1)
, ('Client 2',CONVERT(DATE, '2020-05-10'),0)
, ('Client 2',CONVERT(DATE, '2020-05-11'),1)
) x (Client, RowDate, Value)
)
SELECT
Client
, RowDate
, Value
, CASE
WHEN OnesBefore = DaysInWeek THEN 1
WHEN ZerosBefore = DaysInWeek THEN 0
ELSE PreviousDayValue
END As FinalCalculation
FROM (
-- This set uses windowing to calculate the intermediate values
SELECT
*
-- The count of the days present in the data, as part of the week may be missing we can't assume 7
-- We only count up to this day, so its in line with the other parts of the calculation
, COUNT(RowDate) OVER (PARTITION BY Client, WeekCommencing ORDER BY RowDate) AS DaysInWeek
-- Count up the 1's for this client and week, in date order, up to (and including) this date
, COUNT(IIF(Value = 1, 1, NULL)) OVER (PARTITION BY Client, WeekCommencing ORDER BY RowDate) AS OnesBefore
-- Count up the 0's for this client and week, in date order, up to (and including) this date
, COUNT(IIF(Value = 0, 1, NULL)) OVER (PARTITION BY Client, WeekCommencing ORDER BY RowDate) AS ZerosBefore
-- get the previous days value, or 0 if there isnt one
, COALESCE(LAG(Value) OVER (PARTITION BY Client, WeekCommencing ORDER BY RowDate), 0) AS PreviousDayValue
FROM (
-- This set adds a few simple values in that we can leverage later
SELECT
*
, DATEADD(DAY, -DATEPART(DW, RowDate) + 1, RowDate) As WeekCommencing
FROM Data
) AS DataWithExtras
) AS DataWithCalculations
As you haven't specified your table layout, I don't know what table and field names to use in my example. Hopefully if this is correct you can figure out how to click it in place with what you have - If not, leave a comment
I will note as well, I've made this purposely verbose. If you don't know what the "OVER" clause is, you'll need to do some reading: https://www.sqlshack.com/use-window-functions-sql-server/. The gist is they do aggregations without actually crunching the rows together.
Edit: Adjusted the calculation to be able to account for an arbitrary number of days in the week

Answer 4

Thank you so much to everyone, especially to David and Massimo, which prompted me to restructure the data.
--we join clients and dates each with each and label clients with 'active' or 'inactive'
with a as (
select client, dates
from (select distinct client from dbo.clients) a
cross join (select dates from dates) b
)
, b as (
select date
,1 end active
,client
from clients a
join dbo.dates b on a.id = b.id
)
select client
,a.dates
,isnull(b.active, 0) active
into #tmp2
from a
left join b on a.client= b.client and a.dates = b.dates
--declare variables - for date start and for loop
declare #min_date date = (select min(dates) from #tmp2);
declare #n int = 1
declare #row int = (select count(distinct dates) from #tmp2) --number of the loop iterations
--delete data from the final results
delete from final_results
--fill the table with final results
--run the loop (each iteration = analyse of each 1-week range)
while #n<=#row
begin
with a as (
--run the loop
select client
,max(dates) dates
,sum (case when active = 1 then 1 else null end) sum_active
,sum (case when active = 0 then 1 else null end) sum_inactive
from #tmp2
where dates between dateadd(day, -7 + #n, #min_date) and dateadd(day, -1 + #n, #min_date)
group by client
)
INSERT INTO [dbo].[final_results]
(client
,[dates]
,[final_result])
select client
,dates
,case when sum_active = 7 then 1 --rule A
when sum_inactive = 7 then 0 -- rule B
else
(case when isnull(sum_active, 0) + isnull(sum_inactive, 0) < 7 then 0
else
(select final_result
from final_results b
where b.dates = dateadd(day, -1, a.dates)
and a.client= b.client) end
) end
from a
set #n=#n+1
end
if object_id(N'tempdb..#tmp2', 'U') is not null drop table #tmp2