How do I make the x-axis of plotted timeseries data behave like it does when using datetime values?

I much prefer the zooming behavior of plots created using datetime values to that seen when you plot timeseries data that uses the TimeInfo.StartDate feature. The sample code below illustrates the two approaches. When I zoom in, the bottom axis that uses datetime is much smarter about how times are displayed, offsetting year, month and day to the lower right-hand corner when appropriate.

I have a Simulink model that generates output using timeseries objects and would like the plot method on these objects to have behavior that matches datetime.

I've looked at the properties of each axis but cannot find any significant differences between them. Any suggestions on how to make this happen would be appreciated.

    % Vector of 20 days at 1 minute intervals using MATLAB datenums
    datesDatenum = (737342:(1/1440):737362)';
    % Elapsed time in seconds
    times_sec = (datesDatenum - datesDatenum(1))*86400;

    % Sample values to plot
    values = sind(1:numel(datesDatenum))';

    % Create a timeseries 
    valuesTs = timeseries(values, times_sec);
    valuesTs.TimeInfo.StartDate = datestr(datesDatenum(1));
    valuesTs.TimeInfo.Units = 'seconds';

    % Create dates using datetime
    times_dt = datetime(datesDatenum,'ConvertFrom','datenum');

    figure;  % Create figure and link axes
    % Plot as timeseries
    h(1) = subplot(2,1,1);
    plot(valuesTs); grid on

    % Plot using datetime
    h(2) = subplot(2,1,2);
    plot(times_dt, values);grid on

    linkaxes(h,'x')

[dpb]

Those behavior differences are inherent in the implementation of the two functions; the @timeseries version of plot.m contains some code to generate a "local" Time vector and sets some internal formatting before it calls the @datetime version. You can dig into those internals and see if can ascertain which might possibly be tweaked in a special version but it's pretty deep in the bowels of how TMW has implemented HG2 as to what is actually different and undocumented so good luck.

I'd venture if you want datetime plotting behavior, use and PLOT() a datetime vector will be the only practical way without spending buku's of time trying to figure it out and even then never getting the desired behavior.

>> which plot -all
built-in (C:\ML_R2017\toolbox\matlab\graph2d\plot)
C:\ML_R2017\toolbox\matlab\polyfun\@polyshape\plot.m      % polyshape method
C:\ML_R2017\toolbox\matlab\polyfun\@alphaShape\plot.m     % alphaShape method
C:\ML_R2017\toolbox\matlab\bigdata\@tall\plot.m           % tall method
C:\ML_R2017\toolbox\matlab\graphfun\@graph\plot.m         % graph method
C:\ML_R2017\toolbox\matlab\graphfun\@digraph\plot.m       % digraph method
C:\ML_R2017\toolbox\curvefit\curvefit\@sfit\plot.m        % sfit method
C:\ML_R2017\toolbox\curvefit\curvefit\@cfit\plot.m        % cfit method
C:\ML_R2017\toolbox\signal\signal\@dspdata\plot.m         % dspdata method
C:\ML_R2017\toolbox\stats\classreg\@LinearModel\plot.m    % LinearModel method
C:\ML_R2017\toolbox\matlab\timeseries\@timeseries\plot.m  % timeseries method
>> 

Answer 1

[Doug]

I found that if you set:

valuesTs.TimeInfo.Format='MMM dd'

It will plot your x-axis labels as desired when plotting the timeseries but also throws an error. Not sure why the error.

[Alan Bindemann]

The error is becuase 'MMM' is an invalid format for minutes 'mmm' should be used for months. Its interesting that this places the part of the date below the x-ticks, however, the timeseries method isn't smart enough to move the "common" portion of the ticks as you zoom in and out, which is what I like about the datetime approach.

[Doug]

My Matlab documentation says MMM is Month, abbreviated name. Here's a snip:

When I zoom the x-axis of the timeseries plot I do get the behaviour where the common portion of the ticks move below and to the right. See below. I believe this is what you were looking for, correct? I'm using Matlab version R2021b if that matters.