I was thinking of saving only 6 records in kylin instead of splitting them
outside in daily avg and adding 365 records for each item.  So is there
anyway I can achieve using sql level in kylin or have changes to model to
accomodate above change? Please advice. Thanks

On Wed, Feb 28, 2018 at 5:51 PM, Alberto Ramón <[email protected]>
wrote:

> Sounds like:
> - your minimum granularity for queries are on Weeks, your fact table need
> be on weeks (or less, like days)
> - you will need expand you actual fact table to weeks (or more, days)
> Example use a hive view
> - as extra:  Kylin can't use partition format columns on weeks, the
> minimum es days
>
> Alb
>
> On 28 February 2018 at 21:51, deva namaste <[email protected]> wrote:
>
>> Hello,
>>
>> How would I calculate value for a week while I have bi-monthly values.
>>
>> e.g. Here is my data looks like -
>>
>> Date       -  Value
>> 01/18/2017 -  100
>> 03/27/2017 -  130  (68 Days)
>> 05/17/2017 -  102  (51 Days)
>>
>> I need average value per week, as below. Lets consider between 03/27 and
>> 05/17. So total days between period are 51. so Daily average would be
>> 102/51= 2.04
>>
>> Week4 (Starting March 26, #days = 4) = (4 x 2.04) = 8.16
>> Week1 (Starting Apr 2, #days = 7) = 14.28
>> Week2 (starting Apr 9, #days = 7)= 14.28
>> Week3 (starting Apr 16, #days = 7)= 14.28
>> Week4 (starting Apr 23, #days = 7)= 14.28
>> week5 (Starting Apr 30, #days =7)= 14.28
>> week1 (starting May 7, #days = 7)= 14.28
>> Week2 (starting May 14, #days = 4)= 8.16
>>
>> But as you see that period from 01/18 to 03/27, have 68 days and daily
>> average would be 130/68=1.91
>>
>> So really to get complete week I need 3 days from 130 value and 4 days
>> from 102 value.
>>
>> So real total for that first week would be -
>> Week4 (Starting March 26, #days = 4) = (4x2.04=8.16) + (3x1.91=5.73) =
>> 13.89
>>
>> How would I achieve this in Kylin? Any function? or other method I can
>> use?
>> Just for 6 records for year, I dont want to populate daily records.
>> Thanks
>> Deva
>>
>>
>>
>

Reply via email to