Ranking and obtaining data across moving window - google-bigquery

Question

I have following table -
create table iphone_defects(
product string
,defect string
,qty int64
,fwkyr int64
,fwenddate date
);
insert into iphone_defects values ('iPhone','Glass breakage',100,202112,'2020-09-20');
insert into iphone_defects values ('iPhone','No sound',30,202111,'2020-09-30');
insert into iphone_defects values ('iPhone','Glass breakage',25,202110,'2020-09-06');
insert into iphone_defects values ('iPhone','Audio problem',20,202109,'2020-08-30');
insert into iphone_defects values ('iPhone','No sound',60,202108,'2020-08-23');
insert into iphone_defects values ('iPhone','Empty boxes',30,202107,'2020-08-16');
insert into iphone_defects values ('iPhone','Audio problem',25,202106,'2020-08-09');
Am expecting the following result -
fwkyr refers to Financial Week in a year. I have added in additional column fwenddate basically referring to max date in the financial week of the year.
Basically the ask is to obtain the defect with largest quantity in a 4 week window from the current week. Say for the fwkyr - 202112, the highest defects is for 'Glass breakage' and the total quantity is 100.
This is a static window. My actual use case needs 52 week.
Without the moving window, I know that I can rank and get the data but not sure on how to even approach this problem. Any help?

Answer 1

Per updated question my updated solution gets much longer and changes quite a bit.
I am still not sure if user selects from which week you need another 52 weeks or if you are looking at this calculation from start (week 1) of every year.
I also assume that you have a typo in one of your insert statements when I compare to your desired output table. So I changed it to fit your output table.
1. Create table
create table table.defects(
product string
,defect string
,qty int64
,fwkyr int64
,fwenddate date
);
2. Insert data (adjusted last insert to match your output table)
insert into table.defects values ('iPhone','Glass breakage',100,202112,'2020-09-20');
insert into table.defects values ('iPhone','No sound',30,202111,'2020-09-30');
insert into table.defects values ('iPhone','Glass breakage',25,202110,'2020-09-06');
insert into table.defects values ('iPhone','Audio problem',20,202109,'2020-08-30');
insert into table.defects values ('iPhone','No sound',60,202108,'2020-08-23');
insert into table.defects values ('iPhone','Empty boxes',30,202107,'2020-08-16');
insert into table.defects values ('iPhone','Audio problem',55,202106,'2020-08-09');
3. Query for results
###############################################################################
### start count of weeks since selected first week and
### get number of weeks by desired range
###############################################################################
WITH
get_weeks AS (
SELECT
*,
ROW_NUMBER() OVER(PARTITION BY product ORDER BY fwkyr DESC) AS week_numbering,
SPLIT(CAST(ROW_NUMBER() OVER(PARTITION BY product ORDER BY fwkyr)/4 AS string), '.')[
OFFSET
(0)] AS week_id_0,
FROM
table.defects
ORDER BY
fwkyr DESC
),
###############################################################################
### produce filter column for each window period by offsetting
###############################################################################
get_weeks_consequtive AS (
SELECT
*,
LAG(week_id_0,1) OVER(PARTITION BY product ORDER BY fwkyr DESC) AS week_id_1,
LAG(week_id_0,2) OVER(PARTITION BY product ORDER BY fwkyr DESC) AS week_id_2,
LAG(week_id_0,3) OVER(PARTITION BY product ORDER BY fwkyr DESC) AS week_id_3
FROM
get_weeks ),
###############################################################################
### create tables and calculations per window using filter column where you group by for qty and keep top qty only
###############################################################################
week_id_0 AS (
SELECT
SUM(qty) AS qty,
product,
defect,
week_id
FROM (
SELECT
* EXCEPT(week_id_0,
week_id_1,
week_id_2,
week_id_3),
MAX(fwkyr) OVER() AS week_id
FROM
get_weeks_consequtive
WHERE
week_id_0 = '1' )
GROUP BY
2,
3,
4
ORDER BY
1 DESC
LIMIT
1),
week_id_1 AS (
SELECT
SUM(qty) AS qty,
product,
defect,
week_id
FROM (
SELECT
* EXCEPT(week_id_0,
week_id_1,
week_id_2,
week_id_3),
MAX(fwkyr) OVER() AS week_id
FROM
get_weeks_consequtive
WHERE
week_id_1 = '1' )
GROUP BY
2,
3,
4
ORDER BY
1 DESC
LIMIT
1),
week_id_2 AS (
SELECT
SUM(qty) AS qty,
product,
defect,
week_id
FROM (
SELECT
* EXCEPT(week_id_0,
week_id_1,
week_id_2,
week_id_3),
MAX(fwkyr) OVER() AS week_id
FROM
get_weeks_consequtive
WHERE
week_id_2 = '1' )
GROUP BY
2,
3,
4
ORDER BY
1 DESC
LIMIT
1),
week_id_3 AS (
SELECT
SUM(qty) AS qty,
product,
defect,
week_id
FROM (
SELECT
* EXCEPT(week_id_0,
week_id_1,
week_id_2,
week_id_3),
MAX(fwkyr) OVER() AS week_id
FROM
get_weeks_consequtive
WHERE
week_id_3 = '1' )
GROUP BY
2,
3,
4
ORDER BY
1 DESC
LIMIT
1)
###############################################################################
### union all selected windows
###############################################################################
SELECT
*
FROM
week_id_0
UNION ALL
SELECT
*
FROM
week_id_1
UNION ALL
SELECT
*
FROM
week_id_2
UNION ALL
SELECT
*
FROM
week_id_3
ORDER BY
week_id DESC
get_weeks
get_weeks_consequtive
week_id_1
result
PS ---
I brainstormed this quick per your update perhaps there is a better way and I would be interested in seeing it.
Anyhow, with such lengthy queries I typically produce a python script with text templates for repetitive parts and use a loop to expand repetitive parts to desired lengths by incrementing changing values and inserting them with so called f strings.