How do we assign NaN to multiple table elements most efficiently? Why can't we assign multiple columns at once?

I thought the following code would work to replace specific 0's with NaN in my data table. (Data was not recorded for these treatment fractions.)

 ErrorData = ...
         find(elektapatients.X==0&elektapatients.Y==0&elektapatients.Z==0);
  [elektapatients(ErrorData,:).X elektapatients(ErrorData,:).Y...
                   elektapatients(ErrorData,:).Z]=NaN(length(ErrorData),3);

Instead this results in the error:

 Error using NaN
Too many output arguments.

and I must instead assign each column separately:

   elektapatients(ErrorData,:).X=NaN(length(ErrorData),1); 
   elektapatients(ErrorData,:).Y=NaN(length(ErrorData),1);
   elektapatients(ErrorData,:).Z=NaN(length(ErrorData),1);

Is this the most efficient way? Why didn't MATLAB parse the previous command as 1st column to 1st column, 2nd column to 2nd column, 3rd column to 3rd column?

Answer 1

[the cyclist]

It's the same reason that this does not work:

x = zeros(5,2);
y = zeros(5,2);
z = rand(5,4);

[x, y] = z;

Even though it is "obvious" (to you) that you mean that x should be assigned the first two columns of z, and y should be assigned the other two, it is actually impossible for MATLAB to know that that is what is intended in general. Regardless of the prior dimensions of x and y, MATLAB cannot know that the programmer did not intend for the new x to be 5x3, and the new y to be 5x1.

[Daniel Bridges]

Is there a more elegant solution than to assign values column by column? I thought one of the key advantages of MATLAB was to utilize matrices to simplify code: I thought anything that could be done element by element could be done at once with a vector, and anything done vector by vector could be done at once with a matrix, namely through proper indexing.

Is this not true?

[the cyclist]

Well, I'm not sure I would impugn all of MATLAB, based on one case in which the syntax did not do what you expect. All languages have syntax constraints that make sense in the broader context of the language.

In this case, my impression is that you have tried to shift your thinking back-and-forth from vectors to a matrix, when you could have been in a matrix from the beginning. The problem is actually upstream. Specifically, why do need

elektapatients.X
elektapatients.Y
elektapatients.Z

as separate column vectors at all? Why not a single matrix

elektapatients.XYZ

where I hope it is obvious what I mean. Then, you would have been able to do the natural

elektapatients(ErrorData,:).XYZ] = NaN(length(ErrorData),3);

(or something like that) in one fell swoop.

Instead, you are asking the language to be able to distinguish, syntactically, among

• an Nx3 matrix
• the concatenation of Nx2 and Nx1
• the concatenation of Nx1 and Nx2
• the concatenation of Nx1 and Nx1 and Nx1

and it can get even worse if you realize that maybe you actually had both column and row variables on the left. That's a lot to ask of the language.

Answer 2

[Peter Perkins]

If you're putting NaNs into the same rows of X, Y, and Z, this should do what you want:

elektapatients{errorData,{'X' 'Y' 'Z'}} = NaN % braces, not parens

Hope this helps.