finding distinct customer_id when the part numbers are not one per row but has delimiter “/”

Question

Have a data set similiar to this.

Customer_id PART_N  PART_C           TXN_ID
B123        268888  7902/7900        159
B123        12839   82900/8900       1278
B869        12839   8203/890025/7902 17890
B290        268888  62820/12839      179018

not sure how to combine PART_N and PART_C and find count(distinct customer_id) for each part the same part could be in PART_N or PART_C like part number 12839

I am interested in getting as following table using teradata

Part    COUNT(Distinct Customer id)
268888  2
12839   3
7902    2
7900    1
82900   1
8900    1
8203    1
890025  1
62820   1

if it was just PART_N then it would be straight forward as just one part number is present per row. Unsure how I combine every part number and find how many distinct customer id each one has. If it helps I have all the list of distinct Part numbers in one table say table2.

Answer 1

I cannot not try this code, so see it as pseudocode and sketch of an idea.

SELECT numbers, COUNT(numbers)
FROM
    (SELECT 
        REGEXP_SPLIT_TO_TABLE(             -- B
            CONCAT(PART_N, '/', PART_C),   -- A
            '/'
        ) as numbers
    FROM table) s
GROUP BY numbers                           -- C

A: Concatenation of both columns into one string divided by the delimiter '/'

B: Split string by delimiter

C: Group string parts and count them

http://www.teradatawiki.net/2014/05/regular-expression-functions.html

Answer 2

This is pretty ugly.

First let's split those delimited strings up, using strtok_split_to_table .

create volatile table vt_split as (
select
txn_id,
token as part
from table
    (strtok_split_to_table(your_table.txn_id,your_table.part_c,'/')
    returns (txn_id integer,tokennum integer,token varchar(10))) t
)
with data
primary index (txn_id)
on commit preserve rows;

That will give you all those split apart, with the appropriate txn_id. Then we can union that with the part_n values.

create volatile table vt_merged as  (
select * from vt_split
UNION ALL
select
txn_id,
cast(part_n as varchar(10)) as part
from
vt_foo)
with data
primary index (txn_id)
on commit preserve rows;

Finally, we can join that back to your original table to get the counts of customer by part.

select
    vt_merged.part,
    count (distinct yourtable.customer_id)
from
vt_merged
inner join yourtable
    on vt_merged.txn_id = yourtable.txn_id
group by 1

This could probably done a little bit cleaner, but it should get you what you're looking for.

Answer 3

This is @S-Man's pseudocode as working query:

WITH cte AS
 (
   SELECT Customer_id,
      Trim(PART_N) ||'/' || PART_C AS all_parts
   FROM tab
 )
SELECT
   part, -- if part should be numeric: Cast(part AS INT) 
   Count(DISTINCT Customer_id)
FROM TABLE (StrTok_Split_To_Table(cte.Customer_id, cte.all_parts, '/')
     RETURNS (Customer_id VARCHAR(10), tokennum INTEGER, part VARCHAR(30))) AS t
GROUP BY 1