How to add non-row-sized variable to table?

How can I add a variable to a table if the variable has a different length than the number of rows in the table? One reason to do this if one of the variables is a matrix, here hourlyTemperature,

nDays = 7; % Number of table rows
hoursOfDay = 6:6:24; % Vector which we wish to add  
nHours = length(hoursOfDay); 
hourlyTemperature = rand(nDays, nHours); % Matrix, notionally temperature measured every 6 hours for 7 days
dailyAvgWind = rand(nDays,1); % Vector, notionally daily average wind speed for 7 days
dailyAvgRain = rand(nDays,1); % Vector, notionally daily average rainfall for 7 days
% Make example table
Weather = table(dailyAvgWind, dailyAvgRain, hourlyTemperature, ...
   'VariableNames', {'dailyAvgWind', 'dailyAvgRain', 'hourlyTemperature'});

I want to add hoursOfDay to the table because it is useful for interpreting hourlyTemperature. However since its length differs from the number of table rows (nDays), how do I do this?

Answer 1

[Image Analyst]

Try this:

Weather = table(dailyAvgWind, dailyAvgRain, hourlyTemperature, repmat(hoursOfDay, [nDays, 1]), ...
 'VariableNames', {'dailyAvgWind', 'dailyAvgRain', 'hourlyTemperature', 'HoursOfDay'});

[KAE]

Good idea. You need to transpose the table's matrices to plot them correctly in case that helps someone,

plot(Weather.HoursOfDay', Weather.hourlyTemperature'); % Transposed
xlabel('Hours'); ylabel('Temperature (^oC)');
legend('Day 1', 'Day 2', 'Day 3', 'Day 4', 'Day 5', 'Day 6', 'Day 7')

[KAE]

It would be useful if you could attach metadata to Weather such as

    measurementLocation = 'Colorado';
    instrumentSerialNumber = 1357;

to the table Weather without having to reproduce it for every row. But perhaps that means I should just use a structure.

[Image Analyst]

In many cases structure arrays seem simpler to use. You could try that if you want.

[Peter Perkins]

Put "per array" metadata like that in the table's Description or UserData properties, depending on what you need:

>> Weather.Properties.Description = 'Data from Colorado, Instrument Serial #1357';
>> Weather.Properties.UserData.Location = 'Colorado';
>> Weather.Properties.UserData.InstrumentSerialNumber = 1357;    
>> Weather.Properties
ans = 
  struct with fields:
             Description: 'Data from Colorado, Instrument Serial #1357'
                UserData: [1×1 struct]
          DimensionNames: {'Row'  'Variables'}
           VariableNames: {'dailyAvgWind'  'dailyAvgRain'  'hourlyTemperature'}
    VariableDescriptions: {}
           VariableUnits: {}
                RowNames: {}
>> Weather.Properties.UserData
ans = 
  struct with fields:
                  Location: 'Colorado'
    InstrumentSerialNumber: 1357

Answer 2

[Peter Perkins]

The sampling times are not "one per row", so you're either going to have to replicate them as IA suggests, or store them as metadata, either implicitly in the variable names, or explicitly.

As variable names:

>> Weather = [table(dailyAvgWind, dailyAvgRain), array2table(hourlyTemperature)];
>> Weather.Properties.VariableNames(3:end) = {'hourlyTemperature_0600' 'hourlyTemperature_1200' 'hourlyTemperature_1600' 'hourlyTemperature_2400'}
Weather =
    dailyAvgWind    dailyAvgRain    hourlyTemperature_0600    hourlyTemperature_1200    hourlyTemperature_1600    hourlyTemperature_2400
    ____________    ____________    ______________________    ______________________    ______________________    ______________________
     0.65548         0.82346        0.81472                   0.54688                   0.80028                   0.035712              
     0.17119         0.69483        0.90579                   0.95751                   0.14189                    0.84913              
     0.70605          0.3171        0.12699                   0.96489                   0.42176                    0.93399              
    0.031833         0.95022        0.91338                   0.15761                   0.91574                    0.67874              
     0.27692        0.034446        0.63236                   0.97059                   0.79221                    0.75774              
    0.046171         0.43874        0.09754                   0.95717                   0.95949                    0.74313              
    0.097132         0.38156         0.2785                   0.48538                   0.65574                    0.39223

But presumably you stored the temps as a matrix because you want easy access to them as daily vectors. (With the above, you can get them all as a matrix using braces, i.e. Weather{:,3:end}, but having them as one matrix to begin with may be more convenient.)

As explicit metadata:

>> Weather.Properties.UserData.SamplingTimes = hours(6:6:24);
>> Weather.Properties.UserData
ans = 
  struct with fields:

      SamplingTimes: [6 hr    12 hr    18 hr    24 hr]

By the way, in your original code, this:

'VariableNames', {'dailyAvgWind', 'dailyAvgRain', 'hourlyTemperature', 'HoursOfDay'}

was unnecessary, the table constructor would have picked up those names from the workspace variables.

[KAE]

The capability to store metadata is "make or break" for the utility of tables for me, so thanks for the help. I suggest the documentation page for table is updated to make this information easier to find, i.e. include an example on the main page or otherwise highlight this.