TSQL Averaging over datepart - sql

Question

I have a table with a datetime column in it, consider it an event log for simple, analogous purposes.
I want to produce a report detailing the average number of events that occur at each time of day, to 30 min accuracy.
so the logic is,
get just the time component of each date
round the time to the nearest 30 min window (it can be floored, i.e. 00:29 -> 00:00)
count these (grouped by date)
average all these counts over all days
I also don't want to have any time holes in my data, for example, if nothing occurred in the 00:00 - 00:30 range, i want to report a 0, rather than having a missing row.
How can I achieve this?

Answer 1

WITH TestDates (date) AS (
SELECT CONVERT(DATETIME, '2011-11-15 10:00') UNION ALL
SELECT CONVERT(DATETIME, '2011-11-15 11:31') UNION ALL
SELECT CONVERT(DATETIME, '2011-11-16 10:00')
-- CTE to generate 4 million rows with a sequential integer starting at 0
), GeneratedRows (seq) AS (
SELECT ROW_NUMBER() OVER (ORDER BY N1.number) - 1
FROM master..spt_values AS N1
CROSS JOIN master..spt_values AS N2
WHERE N1.name IS NULL
AND N2.name IS NULL
), RoundedTestDates (date) AS (
SELECT CASE
-- Subtract the minute part
WHEN DATEPART(MINUTE, date) < 25 THEN DATEADD(MINUTE, -1 * DATEPART(MINUTE, date), date)
-- Subtract the minute part, then add an hour
WHEN DATEPART(MINUTE, date) >= 45 THEN DATEADD(HOUR, 1, DATEADD(MINUTE, -1 * DATEPART(MINUTE, date), date))
-- Subtract the minute part, then add an half-hour
ELSE DATEADD(MINUTE, 30, DATEADD(MINUTE, -1 * DATEPART(MINUTE, date), date))
END
FROM TestDates
)
SELECT rounded_date = GeneratedPeriod.date
, ocurrences = COUNT(RoundedTestDates.date)
FROM (SELECT DATEADD(MINUTE, 30 * seq, (SELECT MIN(date) FROM RoundedTestDates))
FROM GeneratedRows
) AS GeneratedPeriod (date)
LEFT JOIN RoundedTestDates
ON GeneratedPeriod.date = RoundedTestDates.date
WHERE GeneratedPeriod.date <= (SELECT MAX(date) FROM RoundedTestDates)
GROUP BY GeneratedPeriod.date
ORDER BY 1

Answer 2

Here is the code you need: (tested in sql2008 and works fine!)
-- Table with the 48 30mins periods of the day
CREATE TABLE #Periods
(
Num INT
)
DECLARE #idt INT
SET #idt = 1
WHILE (#idt <= 48)
BEGIN
INSERT INTO #Periods VALUES (#idt)
SET #idt = #idt + 1
END
--Average of the count for each period on all days.
SELECT DayTable.Num, AVG(CAST(DayTable.DayCount AS DECIMAL))
FROM
( --Total incidents for each interval on each day.
SELECT CAST(FLOOR(CAST(#MyLog.LogDate AS FLOAT)) AS DATETIME) AS DayWithOutTime,
#Periods.Num AS Num,
COUNT(#MyLog.ID) AS DayCount
FROM #Periods LEFT JOIN #MyLog
ON #Periods.Num = (DATEPART(hh, #MyLog.LogDate)*60 + DATEPART(mi,#MyLog.LogDate))/30
GROUP BY CAST(FLOOR(CAST(#MyLog.LogDate AS FLOAT)) AS DATETIME),
#Periods.Num
) AS DayTable
GROUP BY DayTable.Num
DROP TABLE #Periods
Where #NyLog is the table where your datetime is. It shows the count of incidences for each 30min period. The Period 1 is 00:00 -> 00:30 and Period 48 is 23:30 -> 24:00.

Answer 3

In sybase sql is something like this, in sql-server you might need to do some changes but not much :)
create procedure Test #startDay varchar(8), #endDay varchar(8)
as
declare #ocurrence int
declare #numberOfDays int
select #numberOfDays = 0
create table #intervals (
interval_hour int,
interval_min_minute int,
interval_max_minute int,
ocurrences int
)
create table #insertions (
hour int,
minute int
)
declare #hour int, #minute int
select #hour = 0
-- create the intervals
while (#hour <> 24)
begin
insert into #intervals values(#hour,0,29,0)
insert into #intervals values(#hour,30,59,0)
select #hour = #hour + 1
end
while(#startDay <> #endDay)
begin
insert into #insertions
select datepart(hh, *yourcolumn*), datepart(mm, *yourcolumn*) from *yourdb..yourtable* where convert(varchar(8), *yourcolumn*, 112) = #startDay
select #startDay = convert(varchar(8), dateadd(dd, 1, convert(datetime, #startDay, 112)), 112)
select #numberOfDays = #numberOfDays + 1
end
declare cursor1 cursor for
select hour, minute from #insertions
open cursor1
fetch cursor1 into #hour, #minute
while (##sqlstatus=0)
begin
update #intervals
set i.ocurrences = i.ocurrences + 1
from #intervals i
where interval_hour = #hour and #minute between interval_min_minute and interval_max_minute
fetch cursor1 into #hour, #minute
end
close cursor1
select interval_hour 'hour', interval_min_minute 'min minute', interval_max_minute 'max minute', ocurrences,
case when ocurrences > 0 then convert(float, ocurrences) / convert(float, #numberOfDays) else 0 end 'ocurrences average' from #intervals
drop table #intervals
drop table #insertions
go

Answer 4

What I've done is use an auxiliary table of numbers (a 1 column table with number 1 to 1 million) and join to it, adding the value of the number with the dateadd function to the midnight of the date.
since you want 30 minute intervals, then you want to use the dateadd(minute, number*30, yourdate) where number <= 48 (since there are 1440 minutes in a day)/30 = 48 intervals. This will create your time intervals.
Then simply count your occurrences that happen in between the time intervals.