Teradata, Reset When, Partition By, Order By

teradata, reset when, partition by, order by

I was also unsure why this wasn't partitioned by PARTITION BY Y.ACCT_DIM_NB, Y.DAY_TIME_DIM_NB ORDER BY Y.DAY_TIME_DIM_NB, Y.TXN_POSTING_SEQ

Don't know, but this would return a different result (and Y.DAY_TIME_DIM_NB is not needed in ORDER BY because it's already partitioned by it)

Also, is ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW just using the whole partitioned window?

It's exactly the same as ROWS UNBOUNDED PRECEDING, i.e. a syntax variation for a Cumulative Max. The lpartition is ROWS UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING

What does RESET WHEN do in Teradata?

The RESET WHEN is a Teradata extension for dynamically adding partitions, it's a shorter syntax for two (in your case) or three nested OLAP functions:

-- using RESET WHEN
MAX(A.RUN_BAL_AM)
OVER (PARTITION BY A.ACCT_DIM_NB
      ORDER BY A.DAY_TIME_DIM_NB, A.TXN_POSTING_SEQ 
      RESET WHEN A.CS_TXN_CD NOT IN ('072','075','079','107','111','112','139','181','318') 
      ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS  EOD_BAL_AM

-- Same result using Standard SQL
SELECT  
   Max(A.RUN_BAL_AM)
   Over (PARTITION BY A.ACCT_DIM_NB, dynamic_partition
         ORDER BY A.DAY_TIME_DIM_NB, A.TXN_POSTING_SEQ 
         ROWS BETWEEN Unbounded Preceding AND CURRENT ROW) AS  EOD_BAL_AM

FROM 
 ( 
   SELECT
      -- this cumulative sum over 0/1 assigns a new value for each series of rows based on the CASE
      Sum(CASE WHEN A.CS_TXN_CD NOT IN ('072','075','079','107','111','112','139','181','318') THEN 1 ELSE 0 end)
      Over (PARTITION BY A.ACCT_DIM_NB, dynamic_partition
            ORDER BY A.DAY_TIME_DIM_NB, A.TXN_POSTING_SEQ 
            ROWS Unbounded Preceding) AS dynamic_partition
   FROM ...
 ) AS dt

SQL Count reset when binary column equals 0

Assign a grouping to each row. Then use row_number():

select t.*,
       (case when flag = 0 then 0
             else row_number() over (partition by animal, grouping order by order) - 1
        end) as count
from (select t.*,
             sum(case when flag = 0 then 1 else 0 end) over
                 (partition by animal
                  order by order
                  rows between unbounded preceding and current row
                 ) as grouping
      from t
     ) t

The - 1 is because the each sequence of 1 also includes the previous 0. You can also express this as:

             else row_number() over (partition by animal, grouping, flag order by order)

so only the 1s are in the group.

Resetting ROW_NUMBER count based on 3 month gap in date

Teradata supports a proprietary extension to Windowed Aggregates, RESET WHEN, which adds a kind of dynamic partition:

Row_Number() 
Over(PARTITION BY c.src_sbscrbr_id
                , c.src_mbr_sqnc_nbr
                , cl.hlth_srvc_cd 
                , df.serv_prov_id
     ORDER BY  cl.clm_line_srvc_strt_dt
     -- restart the row number when the previous date is more than 3 months ago
     RESET WHEN Min(clm_line_srvc_strt_dt)
                Over (PARTITION BY c.src_sbscrbr_id
                                 , c.src_mbr_sqnc_nbr
                                 , cl.hlth_srvc_cd 
                                 , df.serv_prov_id
                      ORDER BY cl.clm_line_srvc_strt_dt
                      ROWS BETWEEN 1 Preceding AND 1 Preceding) 
               < Add_Months(cl.clm_line_srvc_strt_dt, -3)) AS rncnt

Based on the error message in Gordon's answer your Teradata version doesn't support LAG, yet (must be 16.10+). The MIN is the same as:

                LAG(clm_line_srvc_strt_dt)
                Over (PARTITION BY c.src_sbscrbr_id
                                 , c.src_mbr_sqnc_nbr
                                 , cl.hlth_srvc_cd 
                                 , df.serv_prov_id
                      ORDER BY cl.clm_line_srvc_strt_dt)  

Btw, there's no need to ORDER by the same columns you already use in PARTITION (within a partition it's the same value anyway)

Issue with Ranking / Row Number / Ordering

It looks like this can be solved using a RESET WHEN in your window function:

MAX() OVER(
  PARTITION BY <...> 
  ORDER BY Order_Flow 
  RESET WHEN Storage = 20
)

I believe you can leave out the PARTITION BY if you just want to control the ordering and don't need to do any partitioning. Or just use a constant value, like PARTITION BY 1 or something to that effect.

Documentation:
https://docs.teradata.com/reader/756LNiPSFdY~4JcCCcR5Cw/8uRgqNTevlcmjBfsU3WQsw

Stackoverflow:

teradata, reset when, partition by, order by

---

Related Topics