Custom window in analytic function

Question

SELECT  device_id,
        case_id,
        value,
        fo_value,
        nomination,
        symbol,
        cur_id,
        time,
        data_type,
        COALESCE (
           VALUE,
           LAST_VALUE (
              VALUE IGNORE NULLS
              OVER (PARTITION BY device_id, cur_id, nomination
                    ORDER BY time
                    ROWS UNBOUNDED PRECEDING)
           + SUM(fo_value)
                OVER (PARTITION BY device_id, cur_id, nomination
                      ORDER BY time
                      ROWS UNBOUNDED PRECEDING)
           AS val 
FROM tab  
DEVICE_ID     CASE_ID     VALUE     FO_VALUE     NOMINATION     SYMBOL     CUR_ID     TIME     DATA_TYPE     VAL
8744     4901          -23     20     PLN     1     2009-08-25 11:00:00.000000     F     -48
8744     4901          -2     20     PLN     1     2009-08-25 07:00:00.000000     F     -25
8744     4901     100     -13     20     PLN     1     2009-08-24 23:00:00.000000     I     100
8744     4901          -6     20     PLN     1     2009-08-24 17:00:00.000000     F     45
8744     4901          -5     20     PLN     1     2009-08-24 14:00:00.000000     F     51
8744     4901          -5     20     PLN     1     2009-08-24 11:00:00.000000     F     56VAL column is defined as last VALUE from which there was substracted cummulative sum of FO_VALUE column.
Every record with data_type = 'I' has some number in VALUE column. What I need is to define window in expression that counts sum of FO_VALUE which adds only the values from last NON-NULL VALUE column - currently query substracts all the values in partition.
I need something like window from last NON-NULL in VALUE column row to current row
Desired output should look like this:
DEVICE_ID     CASE_ID     VALUE     FO_VALUE     NOMINATION     SYMBOL     CUR_ID     TIME     DATA_TYPE     VAL
8744     4901          -23     20     PLN     1     2009-08-25 11:00:00.000000     F     62
8744     4901          -2     20     PLN     1     2009-08-25 07:00:00.000000     F     85
8744     4901     100     -13     20     PLN     1     2009-08-24 23:00:00.000000     I     87
8744     4901          -6     20     PLN     1     2009-08-24 17:00:00.000000     F     45
8744     4901          -5     20     PLN     1     2009-08-24 14:00:00.000000     F     51
8744     4901          -5     20     PLN     1     2009-08-24 11:00:00.000000     F     56I'd be very appeciated for any suggestions.

Answer 1

Hi,
Sorry, I don't understand the problem.
In particular, I don't see how you get the vals (56, 51, 45) for the earlier times. It looks like you should get NULL for val in the ealiest 3 rows. Can you explain how you get those numbers?
Is it possible for a partition (device_id, cur_id, nomination) to have 2 or more rows where value is not NULL? If so, please post an example of the results you would want.
Rather than adjusting ROWS ... to get the results you want, it might be easier to adjust the PARTITION BY ...
For example, this gets the results you want on and after the row where value is not NULL:
WITH     got_grp_num
AS
     SELECT  device_id, case_id, value, fo_value, nomination, symbol, cur_id, time, data_type,
          COUNT (value) OVER ( PARTITION BY device_id, cur_id, nomination
                                           ORDER BY         time
                       )         AS grp_num
     FROM     tab
SELECT  device_id, case_id, value, fo_value, nomination, symbol, cur_id, time, data_type,
        LAST_VALUE (value IGNORE NULLS) OVER ( PARTITION BY  device_id, cur_id, nomination, grp_num
                                                  ORDER BY      time
         + SUM (fo_value)           OVER ( PARTITION BY  device_id, cur_id, nomination, grp_num
                                             ORDER BY      time
                              )           AS val 
FROM        got_grp_num  
ORDER BY  device_id, cur_id, nomination,
             time                         DESC
;As you can see, there is a new column in the PARTITION BY clauses that compute val. The new column is grp_num, which has a new value whenever there is a non-null value.
Edited by: Frank Kulash on Aug 28, 2009 1:27 PM