SQL Aggregation logic - Not sure - sql

Question

I am implementing the multiple separate logics between 2 tables by joining them in a view. I need to have minimal number of views with all logics implemented. I am struck with the following issue while implementing and need your expertise.
I had arrived to the base logic of the view as it serves as the base for most of the logics and I am to stick to it;
SELECT Acct_no, max(txn_date),......
FROM ACCT_CRD ac
INNER JOIN TRNSCTN txn ON ( ac.crd_no = txn.crd_no)
GROUP BY ACCT_NO, TO_CHAR(TXN_DATE,'YYYYMM');
Table_name: ACCT_CRD (This table has account and the credit card numbers with UPI on credit card numbers and a single account number can have multiple card_numbers)
Data:
Acct_no | Crd_no | biz_date | Status
--------+--------+------------+--------
acct1 | crd11 | 2015-10-01 | A
--------+--------+------------+--------
acct1 | crd12 | 2015-10-02 | A
--------+--------+------------+--------
acct1 | crd13 | 2015-10-03 | A
Table_name: TRNSCTN (This table has transactions done through the credit cards; data doesn't reflect any actual meaning) Please note that this table has dates with 5 years, for sample, I have took for only 1 month;
Data:
Crd_no | Txn_date | Txn_code | crd_limit | crd_commit
--------+-------------+----------+-------------+------------
crd11 | 2015-10-02 | 10 | 10000 | 9000
--------+-------------+----------+-------------+------------
crd11 | 2015-10-02 | 10 | 10000 | 14000
--------+-------------+----------+-------------+------------
crd11 | 2015-10-02 | 20 | 10000 | 16000
--------+-------------+----------+-------------+------------
crd11 | 2015-10-03 | 20 | 10000 | 12000
--------+-------------+----------+-------------+------------
crd11 | 2015-10-05 | 20 | 10000 | 15000
--------+-------------+----------+-------------+------------
crd12 | 2015-10-03 | 10 | 20000 | 5000
--------+-------------+----------+-------------+------------
crd12 | 2015-10-03 | 20 | 20000 | 22000
--------+-------------+----------+-------------+------------
crd12 | 2015-10-04 | 30 | 20000 | 25000
--------+-------------+----------+-------------+------------
crd12 | 2015-10-04 | 30 | 20000 | 5000
--------+-------------+----------+-------------+------------
crd13 | 2015-10-04 | 30 | 25000 | 10000
Here, in TRNSCTN table, for each card on each day if CRD_COMMIT > CRD_LIMIT, then take the count as 1 even if there are more records with same card_no and txn_date with crd_commit >= crd_limit or crd_commit < crd_limit;
Order is not important on a given day transactions;
SELECT crd_no, txn_date,
MAX(case when crd_commit > crd_limit then 1 else 0 end) day_overlimit_cnt
FROM TRNSCTN group by crd_no, txn_date;
Essentially, the above data in the table transforms to
Crd_no | txn_date | day_overlimit_cnt
-------+-------------+-------------------
crd11 | 2015-10-02 | 1
-------+-------------+-------------------
crd11 | 2015-10-03 | 1
-------+-------------+-------------------
crd11 | 2015-10-05 | 1
-------+-------------+-------------------
crd12 | 2015-10-03 | 1
-------+-------------+-------------------
crd12 | 2015-10-04 | 1
-------+-------------+-------------------
crd13 | 2015-10-04 | 0
Then, I need to find for each card in a given month, how many times it has exceeded the day_overlimit_cnt;
SELECT crd_no, to_char(txn_date,'yyymm') as txn_mnth,
SUM(day_overlimit_cnt) sum_month_ovrlmt from (select crd_no, txn_date,
MAX(case when crd_commit > crd_limit then 1 else 0 end) day_overlimit_cnt
FROM TRNSCTN group by crd_no, txn_date) dt_check
GROUP BY crd_no, to_char(txn_date,'yyymm');
The data from the above query will be
Crd_no | txn_mnth | sum_month_ovrlmt
-------+----------+-----------------
crd11 | 201510 | 3
-------+----------+-----------------
crd12 | 201510 | 2
-------+----------+-----------------
crd13 | 201510 | 0
And then finally find the max(sum_month_ovrlmt) at account level by joining the above one to ACCT_CRD;
SELECT acct_no, MAX(sum_month_ovrlmt) acct_mnth_ovrlmt
FROM ACCT_CRD ac
JOIN (SELECT crd_no, to_char(txn_date,'yyymm') as txn_mnth,
SUM(day_overlimit_cnt) sum_month_ovrlmt FROM (SELECT crd_no, txn_date, MAX(case when crd_commit > crd_limit then 1 else 0 end) day_overlimit_cnt FROM TRNSCTN group by crd_no, txn_date) dt_check GROUP BY crd_no, to_char(txn_date,'yyymm')) dt_dt_check dt on (ac.cr_no = dt.crd_no) GROUP BY acct_no;
Final output:
Acct_no | acct_mnth_ovrlmt
--------+-----------------
acct1 | 3
How to embed the above logic into the following base query; That is how to derive acct_mnth_ovrlmt without affecting the other columns data in the select part.
SELECT Acct_no, max(txn_date),......
FROM ACCT_CRD ac
INNER JOIN TRNSCTN txn ON ( ac.crd_no = txn.crd_no)
GROUP BY ACCT_NO, TO_CHAR(TXN_DATE,'YYYYMM');
Thanks in advance for your time. As a last resort, I will try to embed the above derived code until aggregation of cards into the base query and will try it out.
Greetings Gordon Linoff,
Thank you for your post. I need the distinct conditional count as you mentioned at card_number level. And as account_number can have more than 1 card_number, I need to find out the max(overlimit_cnt) at account_level;
i.e., say if the distinct conditional count is as
crd11 | 3
crd12 | 2
crd13 | 0
As all these card_numbers belong to acct1, need to get the max(overlimit_cnt) of the above card_numbers; i.e.,
acct | 3
I guess I need to again have another join to the same table with group by on different columns as
SELECT Acct_no, max(txn_date),......,MAX(day_overlimit_cnt)
FROM ACCT_CRD ac
INNER JOIN TRNSCTN txn ON ( ac.crd_no = txn.crd_no)
INNER JOIN ( SELECT CRD_NO, TO_CHAR(TXN_DATE,'YYYYMM') AS TXN_DATE_Y,
COUNT(DISTINCT (CASE WHEN crd_commit > crd_limit then TXN_DATE end)) day_overlimit_cnt from TRNSCTN GROUP BY CRD_NO, TO_CHAR(TXN_DATE,'YYYYMM')) TRNSCTN_OVRLMT ON (TRNSCTN.CRD_NO=TRNSCTN_OVRLMT.CRD_NO AND TO_CHAR(TRNSCTN.TXN_DATE,'YYYYMMDD')=TRNSCTN_OVRLMT.TXN_DATE_Y) GROUP BY ACCT_NO, TO_CHAR(TXN_DATE,'YYYYMM');
Can I avoid new join TRNSCTN_OVRLMT and derive above value.

Answer 1

Your logic is a little hard to follow, but I think you just want conditional count distinct:
SELECT Acct_no, max(txn_date),
COUNT(DISTINCT (CASE WHEN crd_commit > crd_limit THEN txn_date END)) as DaysOverLimit
FROM ACCT_CRD ac INNER JOIN
TRNSCTN txn
ON ac.crd_no = txn.crd_no
GROUP BY ACCT_NO, TO_CHAR(TXN_DATE,'YYYYMM');