Plot time-of day histogram for array of datetimes

How do I create a histogram for an array of datetimes showing the counts of times of day from 00:00 to 24:00. I'd like the bin width to be variable between 1 to 60 minutes.

Answer 1

[dpb]

With datetimes, use the duration class for times of day..

[dormant]

Sorry, but I don't understand your answer.

I've only recently started using datetime and have no idea what a class is and how to use it.

[dpb]

So, we don't understand exactly what you have to start with, either. Show us a sample of your data set.

As a rough guess,

x=rescale(randn(1000,1),0,24*60*60-1);      % a set of seconds
d=duration(0,0,x,'format','hh:mm');         % turn into duration
histogram(d)

or

figure
tl=tiledlayout('flow');
nexttile(tl)
d=datetime(2025,1,1,0,0,x,'format','hh:mm');    % turn into datetime
histogram(d)
d=d-datetime(2025,1,1,'Format','hh:mm');        % convert to duration from beginning of day
nexttile(tl)
histogram(d)

gives back the first

Note that if the datetimes include more than one day, using them directly will bin by day and time, not just time; hence the use of a duration instead. The subtraction of the date from the array of values can be vectorized for all; one doesn't have to have a fixed date.

[dormant]

I've got an array of datetimes spanning a period of 5 years.

>> size( datetimeVTs )
ans =
           1        2544

>> datetimeVTs(1:5)
ans = 
  1×5 datetime array

   2020-01-01 07:22:06   2020-01-01 11:35:31   2020-01-02 04:44:33   2020-01-03 12:18:37   2020-01-04 07:19:49

I want to do a histogram that ignores the dates.

[dormant]

It is easier than I thought:

d=timeofday(datetimeVTs);
histogram(d);

[dormant]

Thanks for the help. you pointed me the right way.

[dpb]

No problem -- I should have recalled the timeofday function; subtracting the base datetime for each day is what it does. Seems like maybe when the datetime class was first introduced it wasn't yet there, maybe, is why I fixated on using the duration class; I don't recall precisely. I do remember there being an Answers Q? quite a long time ago about creating some specialized plots that the poster had initially done with datenum required recasting; maybe that recollection got me sidetracked.

Anyways, glad you looked in more depth on your own...

[dpb]

"... have no idea what a class is..."

The object-oriented implementation of data types are classes in MATLAB.

dn=now                                          % get current time as datenum

dn = 7.3989e+05

dt=datetime(dn,'ConvertFrom','datenum')         % convert to a datetime variable

dt = datetime
   04-Oct-2025 16:41:25

t=timeofday(dt)                                 % and get the time of day corresponding

t = duration
   16:41:25

and see what each is in MATLAB

whos

  Name      Size            Bytes  Class       Attributes

  dn        1x1                 8  double                
  dt        1x1                 8  datetime              
  t         1x1                24  duration              

and we see each is a different class; they have different properties and methods specific for what they represent and, in the case of the time-related variables, how they represent time. A datenum is "just" a regular floating point double variable interpreted in a particular way; the integer portion represents the days after the origin reference date and the fractional part the fraction of a day of 24 hours.

datestr(dn)

ans = '04-Oct-2025 16:41:25'

dayn=floor(dn);
datestr(dayn,'dd=mmm-yyyy HH:MM:SS')

ans = '04=Oct-2025 00:00:00'

datestr(dn-dayn,'dd=mmm-yyyy HH:MM:SS')

ans = '00=Jan-0000 16:41:25'

By comparison,

dt-dateshift(dt,'start','day')

ans = duration
   16:41:25

returns the same time within the day as the fractional port of the datenum or timeofday function.

As a general rule, the use of datetime and duration classes is the better choice; in particular plotting as here is greatly simplified for time-related axes that have to be manually configured when using datenums because, as shown above, there is no indication that the particular variable is any different than any other double for the generation of the axes as displaying user-interpretable times. You can now see the advantage of having a class that does impart that knowledge.