How do I calculate the hourly 4 week moving average by weekday? - mysql

Question

Given a table with a datetime column, I want to get the 4 week moving average number of entries per hour with the day of week for each result.
So for instance, between Oct 1st and Oct 13th, I'd like to get back a result that shows the 4 week rolling average for number of rows grouped by hour and dayofweek.
What I have so far gets me the 4 week hourly totals, but not rolling totals:
SELECT
DAYOFWEEK(start_time) as DOW,
date_format( start_time, '%H' ) as 'HOUR',
count( * ) as 'count'
FROM mytable
WHERE start_time >='2017-08-01' and start_time <= '2017-08-29'
GROUP BY DAYOFWEEK(start_time),date_format( start_time, '%H' )

Answer 1

Here is a partially tested approach.
It uses date parameters to ensure consistency of the where clauses. Other parameters are also used to control the hourly bucket (I used 3 in limited testing), and the number of weeks (I used 0 in testing as I had a very small set of rows).
The first subquery is used to produce "ranges" which when joined to the source rows will place those rows into each "rolling n hourly range". Those ranges are defined by using a date_format output YYYYMMDDHH which are strings, and then the data is also forced to this same string format for joining, so if being used on a large table this may cause performance issues (yes, not sargable, I don't like it either).
This solution may be seen working here at SQL Fiddle
Schema Setup:
CREATE TABLE `myTable` (
`id` mediumint(8) unsigned NOT NULL auto_increment,
`start_time` datetime,
PRIMARY KEY (`id`)
) AUTO_INCREMENT=1;
INSERT INTO MyTable
(`start_time`)
VALUES
('2017-08-01 00:01:00'),
('2017-08-01 00:15:00'),
('2017-08-01 00:29:00'),
## more here, 3 rows per hour over a narrow date range
('2017-08-03 08:01:00'),
('2017-08-03 08:15:00'),
('2017-08-03 08:29:00')
;
Query
set #start_time := '2017-08-02';
set #num_hrs := 4; -- controls length of rolling period e.g. 4 hours each
set #num_weeks := 4; -- controls the date date
set #end_time := date_add(#start_time, INTERVAL ((7 * #num_weeks)+1) DAY);
SELECT
DOW
, hour_of_day
, COUNT(*) period_count
, (COUNT(*) * 1.0) / #num_hrs rolling_av
FROM (
## build a set of ranges in YYYYMMDDHH format differing by the wanted number of hours
SELECT
id
, DATE_FORMAT(date_add(start_time, INTERVAL (#num_hrs*-1) HOUR), '%Y%m%d%H') as range_start
, DATE_FORMAT(start_time, '%Y%m%d%H') as range_end
FROM mytable
WHERE start_time >= #start_time and start_time < #end_time
) R
INNER JOIN (
SELECT
start_time
, DAYOFWEEK(start_time) as DOW
, date_format(start_time, '%H' ) as hour_of_day
FROM MyTable
WHERE start_time >= #start_time and start_time < #end_time
) T ON DATE_FORMAT(T.start_time, '%Y%m%d%H') >= R.range_start
AND DATE_FORMAT(T.start_time, '%Y%m%d%H') <= R.range_end
GROUP BY
DOW, hour_of_day
ORDER BY
DOW, hour_of_day
;
Results:
| DOW | hour_of_day | period_count | rolling_av |
|-----|-------------|--------------|------------|
| 4 | 00 | 36 | 12 |
| 4 | 01 | 36 | 12 |
| 4 | 02 | 36 | 12 |
| 4 | 03 | 36 | 12 |
| 4 | 04 | 36 | 12 |
| 4 | 05 | 36 | 12 |
| 4 | 06 | 36 | 12 |
| 4 | 07 | 36 | 12 |
| 4 | 08 | 36 | 12 |
| 4 | 09 | 36 | 12 |
| 4 | 10 | 36 | 12 |
| 4 | 11 | 36 | 12 |
| 4 | 12 | 36 | 12 |
| 4 | 13 | 36 | 12 |
| 4 | 14 | 36 | 12 |
| 4 | 15 | 36 | 12 |
| 4 | 16 | 36 | 12 |
| 4 | 17 | 36 | 12 |
| 4 | 18 | 36 | 12 |
| 4 | 19 | 36 | 12 |
| 4 | 20 | 36 | 12 |
| 4 | 21 | 27 | 9 |
| 4 | 22 | 18 | 6 |
| 4 | 23 | 9 | 3 |