CODEWITHSUNDEEP
sqlsnowflake-cloud-data-platformrank
B.Win
B.Win

Reputation: 43

conditional ranking

I'm trying to rank accounts of a customer by their first payment date. Sometimes the first account they open is never funded and thus has a date '1900-01-01' for "First_pay_date". I want to keep that row of info but do not want to include it in the rank.

Current outcome with code:

CUST ACCT FIRST_PAY_DATE RANK_BY_FIRST_PAY
JOHN H JOHNH1 1900-01-01 NULL
JOHN H JOHNH2 2000-02-25 2
JOHN H JOHNH3 2001-03-21 3
JOHN H JOHNH4 2002-12-01 4

Desired Result:

CUST ACCT FIRST_PAY_DATE RANK_BY_FIRST_PAY
JOHN H JOHNH1 1900-01-01 0
JOHN H JOHNH2 2000-02-25 1
JOHN H JOHNH3 2001-03-21 2
JOHN H JOHNH4 2002-12-01 3 
SELECT
       cust,
       acct,
       first_pay_date,
       CASE
             WHEN first_pay_date <> '1900-01-01' THEN RANK() OVER(PARTITION BY cust
       ORDER BY
             first_pay_date)
       END rank_by_first_pay
FROM
       acct_table
ORDER BY
       first_pay_date ASC

Upvotes: 2

Views: 501

Answers (3)

Simeon Pilgrim
Simeon Pilgrim

Reputation: 25968

In your example output, your RANK function is 0 for the unwanted "1900" date, but what which leads to the simple -1 answer.

So a more logical solution is to making an ADJUISTED_FIRST_PAY you can use NULLIF and use the ORDER NULLS LAST, and then us an IFF to skip that VALUE

long form:

SELECT 
    cust 
    ,acct
    ,NULLIF(first_pay_date,'1900-01-01`) AS clean_date
    ,RANK() OVER (PARTITION BY cust ORDER BY clean_date NULLS LAST) AS ranked
    ,IFF(first_pay_date='1900-01-01`, NULL, ranked) AS answer
FROM
    sample_data

short form:

SELECT 
    cust 
    ,acct
    ,IFF(first_pay_date='1900-01-01`, 
        NULL, 
        RANK() OVER (PARTITION BY cust ORDER BY NULLIF(first_pay_date,'1900-01-01`) NULLS LAST)) AS answer
FROM
    sample_data

which can be pushed to zero like:

SELECT 
    cust 
    ,acct
    ,IFF(first_pay_date='1900-01-01`, 
        0, 
        RANK() OVER (PARTITION BY cust ORDER BY NULLIF(first_pay_date,'1900-01-01`) NULLS LAST)) AS answer
FROM
    sample_data

OR just partition on DATE being the value you want to ignore, and ignore the results, or force them to zero/null

RANK() OVER (PARTITION BY cust, year(first_pay_date)=1900 ORDER BY first_pay_date)

and zero'ed out

SELECT 
    cust 
    ,acct
    ,year(first_pay_date)=1900, 
        0, 
        RANK() OVER (PARTITION BY cust, year(first_pay_date)=1900 ORDER BY first_pay_date)
    ) AS answer
FROM
    sample_data

Upvotes: 1

Adrian White
Adrian White

Reputation: 1804

Another option would be to push the '1900-01-01' to null value and use DENSE_RANK() as recommended above. The benefit is if an account doesn't have the initial 1900-01-01 data point it's not lost. The 1900-01-01 are also nicely placed at the end of the ranking - and if there's more than one instance you'll see them (instead of silently hiding the data error).

with sample_data as (
  select 'JOHN H'CUST ,'JOHNH1' ACCT,'1900-01-01'::date FIRST_PAY_DATE   
union select 'JOHN H'CUST,'JOHNH2' as ACCT,'2000-02-25' as FIRST_PAY_DATE   
union select 'JOHN H'CUST,'JOHNH3' as ACCT,'2001-03-21' as FIRST_PAY_DATE   
union select 'JOHN H'CUST,'JOHNH4' as ACCT,'2002-12-01' as FIRST_PAY_DATE   
union select 'JOBE' CUST,'JOBE1' ACCT,'2000-01-01' FIRST_PAY_DATE    
union select 'JOBE' CUST,'JOBE2' as ACCT,'2000-02-25' as FIRST_PAY_DATE 
union select 'JOBE' CUST,'JOBE3' as ACCT,'2001-03-21' as FIRST_PAY_DATE 
union select 'JOBE' CUST,'JOBE4' as ACCT,'2001-03-21' as FIRST_PAY_DATE )

select 
    CUST 
    ,ACCT
    ,REPLACE(FIRST_PAY_DATE,'1900-01-01','NULL') ADJUSTED_FIRST_PAY_DATE 
    ,DENSE_RANK() OVER (PARTITION BY CUST ORDER BY ADJUSTED_FIRST_PAY_DATE) RANKED_EX_1900
    ,DENSE_RANK() OVER (PARTITION BY CUST ORDER BY FIRST_PAY_DATE)-1 RANKED_MINUS_ONE
FROM
    SAMPLE_DATA

enter image description here

Upvotes: 1

Zakaria
Zakaria

Reputation: 4806

Unless I'm missing something, rank()... - 1 should return the result you want, no need for the case expression:

SELECT
       cust,
       acct,
       first_pay_date,
       RANK() OVER(PARTITION BY cust
       ORDER BY
             first_pay_date) - 1
       rank_by_first_pay
FROM
       acct_table
ORDER BY
       first_pay_date ASC

And depending on the behaviour you want in case of ties, you might want to use dense_rank() instead of rank().

Upvotes: 2

Related Questions