SQL Server Joining Between Two Tables Based on Closest Value - sql-server

Question

I have two sets of data, as shown below.
Table 1:
enter code here
ID Value_1
1 233.67
2 83.28
3 84.49
4 1234.83
Table 2:
NewID Value_3 Value_4
5 NULL 83
6 NULL 85
7 NULL 235
I want to join the two tables in such a way that the resulting data set would look like below.
ID NewID Value_1 Value_2
1 7 233.67 235
2 5 83.28 83
3 6 84.49 85
4 NULL 1234.83 NULL
I know that using the ROUND command would cause future problems. Do any of you know how I could create the above resulting set?

Answer 1

Something like this:
DECLARE #Table1 TABLE
(
Id int,
Value1 float
)
INSERT INTO #Table1
VALUES
(1, 233.67),
(2, 83.28),
(3, 84.49),
(4, 1234.83)
DECLARE #Table2 TABLE
(
NewId int,
Value2 float
)
INSERT INTO #Table2
VALUES
(5, 83),
(6, 85),
(7, 235)
SELECT DISTINCT
t1.Id,
FIRST_VALUE(t2.NewId) OVER (PARTITION BY t1.Id ORDER BY ABS(t1.Value1 - t2.Value2) ASC) AS NewId,
t1.Value1,
FIRST_VALUE(t2.Value2) OVER (PARTITION BY t1.Id ORDER BY ABS(t1.Value1 - t2.Value2) ASC) AS Value2
FROM #Table1 t1
CROSS JOIN #Table2 t2
ORDER BY t1.Id
But you will get a result for ID=4 too.

Answer 2

This approach avoids a cross join, which requires every combination to be tested. It also works with SQL Server 2005 and up. It works by calculating a lower and upper bound between the midpoints of each record in Table2 and then joining on Table1 where Value_1 is between the midpoints. If Value_1 lands right on the boundary this code will round up (choose the higher of Value_4 to match).
--Load Table1
select 1 ID, convert(float,233.67) Value_1 into #Table1
insert into #Table1 select 2, 83.28
insert into #Table1 select 3, 84.49
insert into #Table1 select 4, 1234.83
--Load Table2
select 5 NewID, null Value_3, convert(float,83) Value_4 into #Table2
insert into #Table2 select 6, NULL, 85
insert into #Table2 select 7, NULL, 235
;with cte_Table2 as
(
select *, ROW_NUMBER() over (order by Value_4) OrderNum
from #Table2
)
Select #Table1.ID,
NewTable2.NewID,
#Table1.Value_1,
NewTable2.Value_4 Value_2
from #Table1
full join
(
select Table2.NewID,
Table2.Value_3,
Table2.Value_4,
Table2Prev.Value_4 + (Table2.Value_4 - Table2Prev.Value_4) / 2.0 LowerBound,
Table2.Value_4 + (Table2Next.Value_4 - Table2.Value_4) / 2.0 UpperBound
from cte_Table2 Table2
left join cte_Table2 Table2Prev
on Table2.OrderNum = Table2Prev.OrderNum + 1
left join cte_Table2 Table2Next
on Table2.OrderNum = Table2Next.OrderNum - 1
) NewTable2
on (#Table1.Value_1 < UpperBound or UpperBound is null)
and (#Table1.Value_1 >= LowerBound or LowerBound is null)
order by 1
I noticed that you did not show a match for ID of 4 in your expected output. If you need some kind of range to exclude matches then you would have to add that restriction to the ON conditions of the join. For example you could exclude values that are outside a threshold of 10 by making sure you use a FULL JOIN and add this to the ON conditions:
and abs(#Table1.Value_1-NewTable2.Value_4) < 10.0
It occurs to me though that maybe what you are really trying to do is a type of join where not only is the value in Table1 matched with it's closest value in Table2 but the value in Table2 is also the closest value in Table1. In this case, you would have to build a sub-query for Table1 with the boundary conditions and then check for it's closest match as well. Like this:
;with cte_Table1 as
(
select *, ROW_NUMBER() over (order by Value_1) OrderNum
from #Table1
),
cte_Table2 as
(
select *, ROW_NUMBER() over (order by Value_4) OrderNum
from #Table2
)
Select NewTable1.ID,
NewTable2.NewID,
NewTable1.Value_1,
NewTable2.Value_4 Value_2
from
(
select Table1.ID,
Table1.Value_1,
Table1Prev.Value_1 + (Table1.Value_1 - Table1Prev.Value_1) / 2.0 LowerBound,
Table1.Value_1 + (Table1Next.Value_1 - Table1.Value_1) / 2.0 UpperBound
from cte_Table1 Table1
left join cte_Table1 Table1Prev
on Table1.OrderNum = Table1Prev.OrderNum + 1
left join cte_Table1 Table1Next
on Table1.OrderNum = Table1Next.OrderNum - 1
) NewTable1
full join
(
select Table2.NewID,
Table2.Value_3,
Table2.Value_4,
Table2Prev.Value_4 + (Table2.Value_4 - Table2Prev.Value_4) / 2.0 LowerBound,
Table2.Value_4 + (Table2Next.Value_4 - Table2.Value_4) / 2.0 UpperBound
from cte_Table2 Table2
left join cte_Table2 Table2Prev
on Table2.OrderNum = Table2Prev.OrderNum + 1
left join cte_Table2 Table2Next
on Table2.OrderNum = Table2Next.OrderNum - 1
) NewTable2
on (NewTable1.Value_1 < NewTable2.UpperBound or NewTable2.UpperBound is null)
and (NewTable1.Value_1 >= NewTable2.LowerBound or NewTable2.LowerBound is null)
and (NewTable2.Value_4 < NewTable1.UpperBound or NewTable1.UpperBound is null)
and (NewTable2.Value_4 >= NewTable1.LowerBound or NewTable1.LowerBound is null)
order by 1
Now the records in the tables will only match if the values are the closest match both ways. This will ensure that each record in Table1 will be matched to at most 1 value in Table2. Because it's a full join you can get null values on either side... depending on which table is smaller.
One more thing to mention... this code might not handle duplicate values the way you need. If you have more than one of the same value in either Value_1 or Value_4, it will find a match to one of them but not both. If you want that... then you would have to change your common table expressions to:
;with cte_Table1 as
(
select *, ROW_NUMBER() over (order by Value_1) OrderNum
from (select distinct Value_1 from #Table1) tbl1
),
cte_Table2 as
(
select *, ROW_NUMBER() over (order by Value_4) OrderNum
from (select distinct Value_4 from #Table2) tbl2
)
Then update your sub-queries to only output the values with the boundaries. This would give you best matches with between the unique values. Then you could join to Table1 and Table2 ON Table1.Value_1 = NewTable1.Value1 and Table2.Value_4 = NewTable2.Value_4 to get the rest of the fields.
Certainly some optimizations can be made if you have very large tables... like breaking out some of these sub-queries into indexed temporary tables.