Restructure table by removing NULL values - sql

Question

I have a table in SQL that looks like this:
Customer Product 1999 2000 2001 2002 2003
Smith 51 NULL NULL 15 14 NULL
Jones 14 11 7 NULL NULL NULL
Jackson 13 NULL NULL NULL 3 9
The figures under each year column are amounts, in dollars. Each customer has two consecutive years of amounts, and the rest of the years are zero. I would like to re-structure this table so that instead of wide list of years, it just has two columns Amount-Year1 and Amount-Year2. So it selects the two non-zero years and puts them in those columns, in the correct order. This would greatly reduce the size of my table.
So far I've been able to re-structure it so that there is one amount column and one year column, but I then get multiple rows per customer, which unfortunately I can't have (due to downstream analysis). Can anyone think of a way to get the two Amount-Year columns?
I would like the final table to look like this:
Customer Product Amount_Y1 Amount_Y2
Smith 51 15 14
Jones 14 11 7
Jackson 13 3 9
I don't mind that I lose the information about the specific years, as I can get that from another source. The actual table has data for all years between 1999 and 2018, and there will be further years in the future.
Thanks

Answer 1

Thankfully UNPIVOT removes NULLs anyway, so we can do this with UNPIVOT/ROWNUMBER(),PIVOT:
declare #t table (Customer varchar(15),Product int,[1999] int,
[2000] int,[2001] int,[2002] int,[2003] int)
insert into #T(CUstomer,Product,[1999],[2000],[2001],[2002],[2003]) values
('Smith' ,51,NULL,NULL, 15, 14,NULL),
('Jones' ,14, 11, 7,NULL,NULL,NULL),
('Jackson',13,NULL,NULL,NULL, 3, 9)
;With Numbered as (
select
Customer,Product,Value,
ROW_NUMBER() OVER (PARTITION BY Customer,Product
ORDER BY Year) rn
from
#t t
unpivot
(Value for Year in ([1999],[2000],[2001],[2002],[2003])) u
)
select
*
from
Numbered n
pivot
(SUM(Value) for rn in ([1],[2])) w
Results:
Customer Product 1 2
--------------- ----------- ----------- -----------
Jackson 13 3 9
Jones 14 11 7
Smith 51 15 14

Answer 2

Use COALESCE that will do the job for you. The query is dynamics so that if tomorrow year columns are changed, i.e. removed or added you do not have to change anything.
Sample query: (Assuming table to be table1 and column names to be same as year).
DECLARE #columnsdesc nvarchar(max), #columnsasc nvarchar(max)
SET #columnsdesc = ''
SELECT #columnsdesc = (select + '[' + ltrim(c.Name) + ']' + ','
FROM sys.columns c
JOIN sys.objects o ON o.object_id = c.object_id
WHERE o.type = 'U' and o.Name = 'table1' and c.Name not in ('Customer', 'Product')
ORDER BY c.Name desc for xml path ( '' ))
SET #columnsasc = ''
SELECT #columnsasc = (select + '[' + ltrim(c.Name) + ']' + ','
FROM sys.columns c
JOIN sys.objects o ON o.object_id = c.object_id
WHERE o.type = 'U' and o.Name = 'table1' and c.Name not in ('Customer', 'Product')
ORDER BY c.Name asc for xml path ( '' ))
SELECT #columnsasc = LEFT( #columnsasc,LEN(#columnsasc)-1)
SELECT #columnsdesc = LEFT( #columnsdesc,LEN(#columnsdesc)-1)
DECLARE #sql nvarchar(max)
SET #sql = 'SELECT Customer, Product, COALESCE('+ #columnsasc +') as Amount_Y1,
COALESCE(' + #columnsdesc +' ) as Amount_Y2
FROM Table1'
EXEC(#sql)
If you're dealing with a temporary table, then the code will change slightly:
Test it here: http://rextester.com/MRVR48808
DECLARE #columnsdesc nvarchar(max), #columnsasc nvarchar(max)
SET #columnsdesc = ''
SELECT #columnsdesc = (select + '[' + ltrim(c.Name) + ']' + ','
FROM tempdb.sys.columns c --Changes here
JOIN tempdb.sys.objects o ON o.object_id = c.object_id --Changes here
WHERE o.type = 'U' and o.Name like '#table1%' and c.Name not in ('Customer', 'Product') --Changes here
ORDER BY c.Name desc for xml path ( '' ))
SET #columnsasc = ''
SELECT #columnsasc = (select + '[' + ltrim(c.Name) + ']' + ','
FROM tempdb.sys.columns c --Changes here
JOIN tempdb.sys.objects o ON o.object_id = c.object_id --Changes here
WHERE o.type = 'U' and o.Name like '#table1%' and c.Name not in ('Customer', 'Product') --Changes here
ORDER BY c.Name asc for xml path ( '' ))
SELECT #columnsasc = LEFT( #columnsasc,LEN(#columnsasc)-1)
SELECT #columnsdesc = LEFT( #columnsdesc,LEN(#columnsdesc)-1)
DECLARE #sql nvarchar(max)
SET #sql = 'SELECT Customer, Product, COALESCE('+ #columnsasc +') as Amount_Y1,
COALESCE(' + #columnsdesc +' ) as Amount_Y2
FROM #Table1' --Changes here
EXEC(#sql)

Answer 3

Try using COALESCE as follows : For one field from beginning to end and for the second in the reverse manner.
SELECT Customer,Product, COALESCE([1999],[2000],[2001],[2002],[2003]) as Y1,
COALESCE([2003],[2002],[2001],[2000],[1999]) as Y2
FROM #TEMPDATA

Answer 4

I would do this using cross apply:
select t.customer, t.product, v.Amount_Y1, v.Amount_Y2
from t cross apply
(select max(case when which = 1 then val end) as Amount_Y1,
max(case when which = 2 then val end) as Amount_Y2
from (select val, yr, row_number() over (order by yr) as which
from (values (t.[1999], 1999), (t.[2000], 2000), (t.[2001], 2001),
(t.[2002], 2002), (t.[2003], 2003)
) v(val, yr)
where val is not null
) v