assignment of an object to a vector

Hi all and happy new year!

I have the following simple problem. I have a function say

res=myfunc(a,b,c)
  res=zeros(2,1);
  res(1)=a+log(b)+cos(c);
  res(2)=(b-c)/a;
end

in addition, I have a class say "myclass". When I call the function myfunc with numerical values for a, b and c, the result is as expected. However, when I call myfunc with inputs of class myclass, I get an error saying: "??? In an assignment A(I)=B, the number of elements in B and I must be the same"

clearly, numel(res(1)) is 1 and numel(a+log(b)+cos(c)) is also 1 although res(1) and a+log(b)+cos(c) are of different classes.

A simple workaround is to write the function above differently

res=myfunc2(a,b,c)
  res=[
  a+log(b)+cos(c)
  (b-c)/a
  ];
end

that is, rather that declaring a vector as in the first case, I construct a vector directly. In this case things work fine as well. I just don't know how to make the code work for the first case as well.

Any thoughts?

Thanks, Pat.

Answer 1

[Matt J]

clearly, numel(res(1)) is 1 and numel(a+log(b)+cos(c)) is also 1.

That's not clear at all, since you haven't showed us how LOG, COS, and PLUS were overloaded for myclass in order to obtain the expression a+log(b)+cos(c). We have no idea based, on what you've shown, what this expression produces.

Also, MATLAB does not allow res(1) and res(2) to be of different classes. Therefore when you do

    res(1) = a+log(b)+cos(c)

MATLAB tries to convert a+log(b)+cos(c) from whatever class it is to 'double', the same class as res(2). You haven't mentioned if or how you've overloaded the double() function to accomplish this.

[Patrick Mboma]

Hi Matt, Thank you for the reply. You are right about the overloading of the operations cos, log, plus, etc. You are also right about the fact that the left hand side and the right hand side are of different natures.

Your remarks made me look at the issue in detail and it turns out that the overloaded double CAN return a vector and that is why there is a mismatch in the dimensions. I wanted to take advantage of the speed of a vectorized code. Maybe I should get rid of the overloading of double and just extract the desired output directly from the vector of myclass objects?

[Matt J]

Instead of initializing res to type double by doing

 res=zeros(2,1)

initialize res to a 2x1 vector of type myclass, using the myclass constructor. Rewrite the constructor so that it has this capability.

You could also assign the last element of res first,

    res=myfunc(a,b,c)

      res(2)=(b-c)/a;
      res(1)=a+log(b)+cos(c);

    end

but for this to work, the myclass constructor must be able to run with nargin=0.

[Patrick Mboma]

Getting rid of the overloading of double just made the problem worse.

[Patrick Mboma]

If I initialize res to a myclass object then the function probably won't work for doubles... and I would like to have a procedure that is independent of the input function.

I have a better understanding of the process although my problem remains unsolved. If the left hand side is a double, then matlab calls the double method on the right hand side of the equality. Ideally, the output vector should be resized on the fly to the dimensions of the right hand side. But I guess this is too much asking...

[Matt J]

Then why not use myfunc2? That looks like a perfectly appropriate thing to do anyway, and as a bonus makes things class-independent. Also, what about the 2nd option that I gave you, i.e., start by assigning res(end)? Then assign the remaining elements in any order you wish. Just make sure the constructor of your class can accept nargin=0 arguments.