问题
I have the following table.
Animal Vaccine_Date Vaccine
Cat 2/1/2016 y
Cat 2/1/2016 z
Dog 2/1/2016 z
Dog 1/1/2016 x
Dog 2/1/2016 y
I would like to get the results to be as shown below.
Animal Vaccine_Date Vaccine
Dog 1/1/2016 x
Dog 2/1/2016 y,z
Cat 2/1/2016 y,z
I have the following code which was supplied via my other post at "Combine(concatenate) rows based on dates via SQL"
WITH RECURSIVE recCTE AS
(
SELECT
animal,
vaccine_date,
CAST(min(vaccine) as VARCHAR(50)) as vaccine, --big enough to hold concatenated list
cast (1 as int) as depth --used to determine the largest/last group_concate (the full group) in the final select
FROM TableOne
GROUP BY 1,2
UNION ALL
SELECT
recCTE.animal,
recCTE.vaccine_date,
trim(trim(recCTE.vaccine)|| ',' ||trim(TableOne.vaccine)) as vaccine,
recCTE.depth + cast(1 as int) as depth
FROM recCTE
INNER JOIN TableOne ON
recCTE.animal = TableOne.animal AND
recCTE.vaccine_date = TableOne.vaccine_date and
TableOne.vaccine > recCTE.vaccine
WHERE recCTE.depth < 5
)
--Now select the result with the largest depth for each animal/vaccine_date combo
SELECT * FROM recCTE
QUALIFY ROW_NUMBER() OVER (PARTITION BY animal,vaccine_date ORDER BY depth desc) =1
But this results in the following.
Animal Vaccine_Date vaccine depth
Cat 2/1/2016 y,z,z,z,z 5
Dog 1/1/2016 x 1
Dog 2/1/2016 y,z,z,z,z 5
The "z" keeps repeating. This is because the code is saying anything greater than the minimum vaccine. To account for this, the code was changed to the following.
WITH RECURSIVE recCTE AS
(
SELECT
animal,
vaccine_date,
CAST(min(vaccine) as VARCHAR(50)) as vaccine, --big enough to hold concatenated list
cast (1 as int) as depth, --used to determine the largest/last group_concate (the full group) in the final select
vaccine as vaccine_check
FROM TableOne
GROUP BY 1,2,5
UNION ALL
SELECT
recCTE.animal,
recCTE.vaccine_date,
trim(trim(recCTE.vaccine)|| ',' ||trim(TableOne.vaccine)) as vaccine,
recCTE.depth + cast(1 as int) as depth,
TableOne.vaccine as vaccine_check
FROM recCTE
INNER JOIN TableOne ON
recCTE.animal = TableOne.animal AND
recCTE.vaccine_date = TableOne.vaccine_date and
TableOne.vaccine > recCTE.vaccine and
vaccine_check <> recCTE.vaccine_check
WHERE recCTE.depth < 5
)
--Now select the result with the largest depth for each animal/vaccine_date combo
SELECT * FROM recCTE
QUALIFY ROW_NUMBER() OVER (PARTITION BY animal,vaccine_date ORDER BY depth desc) =1
However, this resulted in the following.
Animal Vaccine_Date vaccine depth vaccine_check
Cat 2/1/2016 y 1 y
Dog 1/1/2016 x 1 x
Dog 2/1/2016 y 1 y
What is missing in the code to get the desired results of the following.
Animal Vaccine_Date Vaccine
Dog 1/1/2016 x
Dog 2/1/2016 y,z
Cat 2/1/2016 y,z
回答1:
Hmmm. I don't have Teradata on hand but this is a major shortcoming in the project (in my opinion). I think this will work for you, but it might need some tweaking:
with tt as (
select t.*,
row_number() over (partition by animal, vaccine_date order by animal) as seqnum
count(*) over (partition by animal, vaccine_date) as cnt
),
recursive cte as (
select animal, vaccine_date, vaccine as vaccines, seqnum, cnt
from tt
where seqnum = 1
union all
select cte.animal, cte.dte, cte.vaccines || ',' || t.vaccine, tt.seqnum, tt.cnt
from cte join
tt
on tt.animal = cte.animal and
tt.vaccine_date = cte.vaccine_date and
tt.seqnum = cte.seqnum + 1
)
select cte.*
from cte
where seqnum = cnt;
回答2:
If your Teradata Database version is 14.10 or higher it supports XML data type. This also means that XMLAGG function is supported which would be useful for your case and would let you avoid recursion.
Check if XMLAGG function exists, which is installed with XML Services as an UDF:
SELECT * FROM dbc.FunctionsV WHERE FunctionName = 'XMLAGG'
If it does, then the query would look like:
SELECT
animal,
vaccine_date
TRIM(TRAILING ',' FROM CAST(XMLAGG(vaccine || ',' ORDER BY vaccine) AS VARCHAR(10000)))
FROM
tableone
GROUP BY 1,2
I have no way of testing this atm, but I believe this should work with possibility of minor tweaks.
回答3:
I was able to get the desired results with the following SQL. This doesn't seem very efficient at all and is not dynamic. However, I can add extra sub querys as needed to combine more vaccines by animal by date.
select
qrya.animal
,qrya.vaccine_date
,case when qrya.vac1 is not null then qrya.vac1 else null end ||','||case when qrya.animal=qryb.animal and qrya.vaccine_date=qryb.vaccine_date then qryb.Vac2 else 'End' end as vaccine_List
from
(
select
qry1.Animal
,qry1.Vaccine_Date
,case when qry1.Vaccine_Rank = 1 then qry1.vaccine end as Vac1
from
(
select
animal
,vaccine_date
,vaccine
,row_number() over (partition by animal,vaccine_date order by vaccine) as Vaccine_Rank
from TableOne
) as qry1
where vac1 is not null
group by qry1.Animal,
qry1.Vaccine_Date
,case when qry1.Vaccine_Rank = 1 then qry1.vaccine end
) as qrya
join
(
select
qry1.Animal
,qry1.Vaccine_Date
,case when qry1.Vaccine_Rank = 2 then qry1.vaccine end as Vac2
from
(
select
animal
,vaccine_date
,vaccine
,row_number() over (partition by animal,vaccine_date order by vaccine) as Vaccine_Rank
from TableOne
) as qry1
where vac2 is not null
group by qry1.Animal,
qry1.Vaccine_Date
,case when qry1.Vaccine_Rank = 2 then qry1.vaccine end
) as qryb
on qrya.Animal=qryb.Animal
来源：https://stackoverflow.com/questions/39575359/combining-rows-in-sql-viia-recursive-query