Hive windowing query - hive

Question

I have a base Hive table with following schema:
And I want the below output:
So basically, grouping on all columns, and calculating the count distinct Encounters in that month and last 3 months (including that month).
For example, for DischargeMonthYear Jan-2018, num_discharges_last_30_days would be patients discharged in Jan-2018 (3) and num_discharges_last_90_days would be patients discharged in Nov-17, Dec-17 and Jan-18. Since there is no data before Jan-18 in this case, both counts would be the same.
Similarly for Mar-18, num_discharges_last_90_days should include counts for Jan, Feb and Mar-18 months (3+2+2 = 7).
For Jun-18, since we have no data for Apr and May-18, it should include counts only for Jun-18 and NOT got to the previous group/partition.
I have the below query that gives me the correct total for num_discharges_last_90_days till Jun-18 but does not follow the grouping of earlier columns and for Jul-18 it also includes Jun-18 totals which should not be the case since the region is different.
If I add a PARTITION BY region (and others) clause for it, num_discharges_last_90_days is correct for Jul-18 now, but incorrect for Jun-18 since it includes the Feb and Mar-18 totals.
`
DROP TABLE IF EXISTS Encounter;
CREATE TEMPORARY TABLE Encounter
(
Encounter_no int,
Admit_date date,
discharge_date date,
region varchar(50),
Facilityname varchar(50),
Payertype varchar(10),
Payernamme varchar(20),
patient_type varchar(10)
);
INSERT INTO Encounter
select 12345, '2018-01-01', '2018-01-05', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12346, '2018-01-02', '2018-01-06', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12347, '2018-01-03', '2018-01-07', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12348, '2018-02-04', '2018-02-08', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12349, '2018-02-05', '2018-02-09', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12350, '2018-03-06', '2018-03-10', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12351, '2018-03-07', '2018-03-11', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12352, '2018-06-08', '2018-06-12', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12353, '2018-06-09', '2018-06-13', 'Midwest', 'ABC', 'MCR', 'MCR123', 'IP' union all
select 12354, '2018-07-10', '2018-07-14', 'NorthEast', 'ABC', 'MCR', 'MCR123', 'IP'
;
--SELECT from_unixtime(unix_timestamp(e.discharge_date, 'yyyy-MM-dd'),'MM') AS `Discharge_Month` FROM Encounter e
--Below CTE is used to get all month numbers
WITH R AS
(
SELECT '01' AS MonthNum
UNION ALL SELECT '02'
UNION ALL SELECT '03'
UNION ALL SELECT '04'
UNION ALL SELECT '05'
UNION ALL SELECT '06'
UNION ALL SELECT '07'
UNION ALL SELECT '08'
UNION ALL SELECT '09'
UNION ALL SELECT '10'
UNION ALL SELECT '11'
UNION ALL SELECT '12'
)
SELECT * FROM
(
--Perform a left join on CTE with your query to get all months
SELECT
R.MonthNum,
e.region,
e.facilityname,
from_unixtime(unix_timestamp(e.discharge_date, 'yyyy-MM-dd'),'MMM-yyyy') AS Discharge_Month,
e.Payertype,
e.Payernamme,
e.patient_type,
CASE WHEN COALESCE(e.region, '') <> ''
THEN COUNT(1)
ELSE 0
END
as num_discharges_last_30_days,
SUM(
CASE WHEN COALESCE(e.region, '') <> ''
THEN COUNT(1)
ELSE 0
END
)
OVER (ORDER BY R.MonthNum
ROWS BETWEEN 2 PRECEDING AND CURRENT ROW
) as num_discharges_last_90_days
FROM R
LEFT JOIN Encounter e
ON R.MonthNum = from_unixtime(unix_timestamp(e.discharge_date, 'yyyy-MM-dd'),'MM')
GROUP BY
R.MonthNum,
e.region,
e.facilityname,
from_unixtime(unix_timestamp(e.discharge_date, 'yyyy-MM-dd'),'MMM-yyyy'),
e.Payertype,
e.Payernamme,
e.patient_type
) A
WHERE A.region IS NOT NULL
;
`

Answer 1

My colleague cracked the question using the below query. It needed a self-join and CASE & WHERE clauses to only count the last 3 months calculation.
WITH CTE AS (
SELECT a.region,a.facilityname,a.payertype,a.payernamme,a.patient_type, LAST_DAY(a.discharge_date) AS month_year, COUNT(encounter_no) AS measure_1
FROM Encounter AS a
GROUP BY a.region,a.facilityname,a.payertype,a.payernamme,a.patient_type, LAST_DAY(a.discharge_date)
)
-- SELECT * FROM CTE AS a;
SELECT a.region,a.facilityname,a.payertype,a.payernamme,a.patient_type, a.month_year, MAX(a.measure_1) AS measure_1,
SUM(IF(b.month_year IS NULL, a.measure_1, b.measure_1)) AS measure_2
FROM CTE AS a
LEFT JOIN CTE AS b
ON a.region = b.region
AND a.facilityname = b.facilityname
AND a.payertype = b.payertype
AND a.payernamme = b.payernamme
AND a.patient_type = b.patient_type
WHERE ( b.month_year BETWEEN add_months(a.month_year, -2) AND a.month_year
OR b.month_year IS NULL)
GROUP BY a.region,a.facilityname,a.payertype,a.payernamme,a.patient_type, a.month_year;