Talk:Averages/Mean time of day

It seems redundant to have a task that is exactly the same as Averages/Mean angle except that 360 degrees is replaced with 24 hours. --Spoon! 08:57, 12 July 2012 (UTC)

A bit. I agree, except I suspect most of the code for this task will involve going to and from the printable time format. I added it to the "Date and time" category. —Sonia 17:39, 21 September 2012 (UTC)

You have to factor in minutes and hours here, so the task may not be completely redundant. Markhobley 19:32, 11 February 2013 (UTC)

I think this is ready to promote to task. Markhobley 19:32, 11 February 2013 (UTC)

TCL and rounding

TCL has a different output than others, of 23:47:44 instead of 23:47:43. It seems to be somewhere in the calculation of the mean time rather than in the split into HH:MM:SS as I tried the TCL way in my python - inserting the two lines before the return statement below and got 43 seconds and not 44 again. <lang python>def mean_time(times):

   t = (time.split(':') for time in times)
   seconds = ((float(s) + int(m) * 60 + int(h) * 3600) 
              for h, m, s in t)
   day = 24 * 60 * 60
   to_angles = [s * 360. / day for s in seconds]
   mean_as_angle = mean_angle(to_angles)
   mean_seconds = mean_as_angle * day / 360.
   if mean_seconds < 0:
       mean_seconds += day
   h, m = divmod(mean_seconds, 3600)
   m, s = divmod(m, 60)
   a = mean_seconds
   print("%02d:%02d:%02d" % (a / 60 / 60 % 24, a / 60 % 60, a % 60))
   return '%02i:%02i:%02i' % (h, m, s)</lang>

--Paddy3118 (talk) 05:57, 2 July 2013 (UTC)

Was a rounding issue; int() rounds to zero whereas round() rounds to nearest. –Donal Fellows (talk) 13:05, 28 August 2013 (UTC)

I'm not convinced that whole methodology is correct in the first place. For example, when I try an alternative mechanism for time averaging (with times being either “pre” or “post” midnight so as to minimise the deltas): <lang tcl>% set t [clock scan 23:00:17 -base 0] 79217 % incr t [clock scan 23:40:20 -base 0] 160837 % incr t [clock add [clock scan 00:12:45 -base 0] 1 day] 244402 % incr t [clock add [clock scan 00:17:19 -base 0] 1 day] 328241 % expr $t/4 82060 % clock format [expr $t/4] -format %H:%M:%S 23:47:40</lang> As you can see, I get a different answer (several seconds out) and that's using exact arithmetic. (You might get different intermediate values — they're local-timezone-dependant without the -gmt true option — but the final formatted result should be the same.) –Donal Fellows (talk) 12:59, 28 August 2013 (UTC)

I agree. The mean of midnight - 3583 seconds, midnight - 1180 seconds, midnight + 765 seconds, and midnight + 1039 seconds is midnight - 739.75 seconds, or 23:47:40 to the nearest second. All the solutions so far — except for the JavaScript (which doesn't use the specified method) and the Scala (whose actual result isn't shown) — are claiming 23:47:43. --Nig (talk) 12:18, 4 April 2020 (UTC)