Skip to content
Advertisement

Create Missing Data Hive SQL

I have a table that has an activity date of when things change such as

2020-08-13  123 Upgrade
2020-08-17  123 Downgrade
2020-08-21  123 Upgrade

Basically this in relation to a line there are 3 activities happening on this account. They have a basic account then they downgrade but then they upgrade again

I would like to have these happen in steps such as

2020-08-13  123 Upgrade1
2020-08-14  123 Upgrade1
2020-08-15  123 Upgrade1
2020-08-16  123 Upgrade1
2020-08-17  123 Downgrade1
2020-08-18  123 Downgrade1
2020-08-19  123 Downgrade1
2020-08-20  123 Downgrade1
2020-08-21  123 Upgrade2
.
.
.
2020-09-09 123 Upgrade2

Then I would like to partition them by their activity and see this in the end results so I can calculate how many users stayed in their downgraded state for more than 30 days to see their behavior compared to a upgraded change.

2020-08-13  123 Upgrade1. 1
2020-08-14  123 Upgrade1. 2
2020-08-15  123 Upgrade1. 3
2020-08-16  123 Upgrade1. 4
2020-08-17  123 Downgrade1. 1
2020-08-18  123 Downgrade1. 2
2020-08-19  123 Downgrade1. 3
2020-08-20  123 Downgrade1. 4
2020-08-21  123 Upgrade2.   1
.
.
.
2020-09-09 123 Upgrade2.  18

I have tried doing Coalesce then row_num but I can’t wrap my head around how to partition out each activity based on when they changed their account status.

Advertisement

Answer

Generate rows using posexplode(split(space(datediff(next_date,activity_date)-1),’ ‘)). Calculate new_group flag when previous activity<>current activity. Use analytic sum() to calculate group(partition) number. See comments in the code:

with mydata as (
select stack(3,
'2020-08-13', 123, 'Upgrade',
'2020-08-17', 123, 'Downgrade',
'2020-08-21', 123, 'Upgrade'
) as (activity_date, account, activity)
)

--calculate row_number in account, activity
select activity_date, account, activity, activity_partition,
       row_number() over(partition by account, activity_partition order by activity_date ) activity_partition_rn,
       count(*) over(partition by account, activity_partition )     days_on_activity
from
(--Calculate partition
select activity_date, account, activity,
      concat(activity,
             sum(new_group_flag) over(partition by account,  activity order by activity_date rows between unbounded preceding and current row)
            ) activity_partition
from
(--Calculate new group flag
select activity_date, account, activity, 
        case when lag(activity) over (partition by account order by activity_date) = activity then 0 else 1 end as new_group_flag
from
( --generate Date range
select  date_add(activity_date,i) as activity_date, account, activity
from
( --Get next_date to generate date range
select activity_date, account, activity,
       lead(activity_date,1, activity_date) over (partition by account order by activity_date) next_date    
  from mydata d  
) s lateral view posexplode(split(space(datediff(next_date,activity_date)-1),' ')) e as i,x --generate rows
)s
)s
)s
order by activity_date;

Result:

activity_date    account    activity    activity_partition    activity_partition_rn    days_on_activity
2020-08-13            123    Upgrade      Upgrade1                1                         4
2020-08-14            123    Upgrade      Upgrade1                2                         4
2020-08-15            123    Upgrade      Upgrade1                3                         4
2020-08-16            123    Upgrade      Upgrade1                4                         4
2020-08-17            123    Downgrade    Downgrade1              1                         4
2020-08-18            123    Downgrade    Downgrade1              2                         4
2020-08-19            123    Downgrade    Downgrade1              3                         4
2020-08-20            123    Downgrade    Downgrade1              4                         4
2020-08-21            123    Upgrade      Upgrade2                1                         1
User contributions licensed under: CC BY-SA
9 People found this is helpful
Advertisement