Changing class of the variables in the table - I can't get it to work without a loop

See attached sample data.

It contains a table with recorded data in the following fashion:

• Column 1: duration
• Column 2: Force 1
• Column 3: Force 2
• Column 4: Speed

Every four column a new cycle is recorded.

The problem is that some of the columns are recorded as 'strings' and I want to convert them to 'double'.

I've tried a few approaches, one of them works - WAY 1 - but it involves using 'for' loop, which I want to avoid using. Ideally, I would like the WAY 2 get to work, but I don't know how. Any suggestions?

load t_sample.mat;

% Objective: convert all of the 'string' columns into 'double'

strColumnIndex = 1:2:size(t_sample,2);

copyONE   = t_sample;
copyTWO   = t_sample;
copyTHREE = t_sample;
copyFOUR  = t_sample;

% 1st WAY: 
% it works but I don't want to use the loop

for i = strColumnIndex
   copyONE.(i) = cast(copyONE.(i),'double');
end

% 2nd WAY:
% It doesn't work.

tempVariable =  cast(copyTWO{:,strColumnIndex},'double');
copyTWO{:,strColumnIndex} = tempVariable;

% 3rd WAY
% It doesn't work.

varNames = copyTHREE.Properties.VariableNames(strColumnIndex);
copyTHREE{:,varNames} = cast(copyTHREE{:,varNames},'double');

% 4th WAY
% Produces error

varNames1 = copyFOUR.Properties.VariableNames(strColumnIndex);
copyFOUR.varNames1 = cast(copyFOUR.varNames1,'double');

I've looked up ways of accessing data in the tables:

https://uk.mathworks.com/help/matlab/matlab_prog/access-data-in-a-table.html

And to me, all 4 ways presented in the code are equivalent. What am I missing?

Answer 1

[Peter Perkins]

Pawel, the only way to change the type of a variable in a table is to overwrite it. Assigning with t(:,vars) = ... or t{:,vars} = ... are assignments to specific elements of the table. These syntaxes have to account for the possibility that the rows subscript might also have been, for example, 2:end-1, and you certainly would not want t(2:end-1) = ... to change the type of those variables.

Your 4th way is an error simply because you must follow the dot with a reference to only one variable. Which is to say, using dot subscripting to change the type of more than one variable is going to need a loop, as in your 1st way.

It's also possible to do something like this:

strVars = <a logical expression>
t1 = t(:,~strVars);
t2 = varfun(@(s)cast(s,'double'),t,'InputVariables',strVars];
t = [t1 t2];

but you'll also have to put things back to their original order.

Is there a specific reason why that won't work for you? Generally avoiding loops in MATLAB is a good idea, but this is not really one of those cases. If it's a readability thing, you can just make a function called something like castTextVarsToNumeric that encapsulates the loop.

[Pawel Jastrzebski]

Hello Peter,

thank you for your reply.

My precise reason for avoiding the loop is like you said - trying to stick to the best practice. It's that, and also the fact that I use loops somewhere else in my code to process the data - to extract precisely those 4 columns from the raw cycle data file (each cycle is in the separate file).

I simply don't want to use too many loops if I don't have to.