SQL Server retrieving multiple columns with rank 1 - sql

Question

I will try to describe my issue as clearly as possible.
I have a dataset of unique 1000 clients, say ##temp1.
I have another dataset which holds the information related to the 1000 clients from ##temp1 across the past 7 years. Lets call this dataset ##temp2. There are 6 specific columns in this second dataset (##temp2) that I am interested in, lets call them column A, B, C, D, E, F. Just for information, the information that columns A, C, E hold is year of some form in data type float (2012, 2013, 2014..) and the information that columns B, D, F hold is ratings of some form (1,2,3,..upto 5) in data type float. Both year and ratings columns have NULL values which I have converted to 0 for now.
My eventual goal is to create a report which holds the information for the 1000 clients in ##temp1, such that each row should hold the information in the following form,
ClientID | ClientName | ColA_Latest_Year1 | ColB_Corresponding_Rating_Year_1 | ColC_Latest_Year2 | ColD_Corresponding_Rating_Year_2 | ColE_Latest_Year3 | ColF_Corresponding_Rating_Year3.
ColA_Latest_Year1 should hold the latest year for that particular client from dataset ##temp2 and ColB_Corresponding_Rating_Year_1 should hold the rating from Column B corresponding to the year pulled in from Column A. Same goes for the other columns.
The approach which I have taken so far, was,
Create ##temp1 as needed
Create ##temp2 as needed
##temp1 LEFT JOIN ##temp2 on client ids to retrieve the year and rating information for all the clients in ##temp1 and put all that information in ##temp3. There will be multiple rows for every client in ##temp3 because the data in ##temp3 is for multiple years.
Ranked the year column (B,D,F) partition by client_ids and put in in ##temp4,
What I have now is something like this,
Rnk_A | Rnk_C | Rnk_F | ColA | ColB | ColC | ColD | ColE | ColF | Client_id | Client_name
2 | 1 | 1 | 0 | 0 | 0 | 0 | 2014 | 1 | 111 | 'ABC'
1 | 2 | 1 | 2012 | 1 | 0 | 0 | 0 | 0 | 111 | 'ABC'
My goal is
Rnk_A | Rnk_C | Rnk_F | ColA | ColB | ColC | ColD | ColE | ColF | Client_id | Client_name
1 | 1 | 1 | 2012| 1 | 0 | 0 | 2014| 1 | 111 | 'ABC'
Any help is appreciated.

Answer 1

This answer assumes you don't have any duplicates per client in columns A, C, E. If you do have duplicates you'd need to find a way to differentiate them and make the necessary changes.
The hurdle that you've failed to overcome in your attempt (as described) is that you're trying to join from temp1 to temp2 only once for lookup information that could come from 3 distinct rows of temp2. This cannot work as you hope. You must perform separate joins for each pair [A,B] [C,D] and [E,F]. The following demonstrates a solution using CTEs to derive the lookup data for each pair.
/********* Prepare sample tables and data ***********/
declare #t1 table (
ClientId int,
ClientName varchar(50)
)
declare #t2 table (
ClientId int,
ColA datetime,
ColB float,
ColC datetime,
ColD float,
ColE datetime,
ColF float
)
insert into #t1
select 1, 'Client 1' union all
select 2, 'Client 2' union all
select 3, 'Client 3' union all
select 4, 'Client 4'
insert into #t2
select 1, '20001011', 1, '20010101', 7, '20130101', 14 union all
select 1, '20040101', 4, '20170101', 1, '20120101', 1 union all
select 1, '20051231', 0, '20020101', 15, '20110101', 1 union all
select 2, '20060101', 2, NULL, 15, '20110101', NULL union all
select 2, '20030101', 3, NULL, NULL, '20100101', 17 union all
select 3, NULL, NULL, '20170101', 42, NULL, NULL
--select * from #t1
--select * from #t2
/********* Solution ***********/
;with MaxA as (
select ROW_NUMBER() OVER (PARTITION BY t2.ClientId ORDER BY t2.ColA DESC) rn,
t2.ClientId, t2.ColA, t2.ColB
from #t2 t2
--where t2.ColA is not null and t2.ColB is not null
), MaxC as (
select ROW_NUMBER() OVER (PARTITION BY t2.ClientId ORDER BY t2.ColC DESC) rn,
t2.ClientId, t2.ColC, t2.ColD
from #t2 t2
--where t2.ColC is not null and t2.ColD is not null
), MaxE as (
select ROW_NUMBER() OVER (PARTITION BY t2.ClientId ORDER BY t2.ColE DESC) rn,
t2.ClientId, t2.ColE, t2.ColF
from #t2 t2
--where t2.ColE is not null and t2.ColF is not null
)
select t1.ClientId, t1.ClientName, a.ColA, a.ColB, c.ColC, c.ColD, e.ColE, e.ColF
from #t1 t1
left join MaxA a on
a.ClientId = t1.ClientId
and a.rn = 1
left join MaxC c on
c.ClientId = t1.ClientId
and c.rn = 1
left join MaxE e on
e.ClientId = t1.ClientId
and e.rn = 1
If you run this you may notice some peculiar results for Client 2 in columns C and F. This is because (as per your question) there may be some NULL values. ColC date is "unknown" and ColF rating is "unknown".
My solution preserves NULL values instead of converting them to zeroes. This allows you to handle them explicitly if you so choose. I commented out lines in the above query that could be used to ignore NULL dates and ratings if necessary.