How to make a two-dimensional mask act on a three-dimensional array?

Here's a dummy three-dimensional array 'B' to illustrate the problem:

A = 9* ones (6,4,3); % create a 3 dimensional array
A(4,3,2)=NaN; % add a NaN
A(3,2,1)=NaN; % add another NaN;
B = A;

B is a 6x4x3 array with two NaN elements.

Wherever a NaN appears in array B, I want a NaN to overwrite all elements in the corresponding first dimension. So, in the example above, I want to convert B so that my output is

B(:,:,1) =

     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9


B(:,:,2) =

     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9


B(:,:,3) =

     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9

The script below appears to work but is not very elegant. Is there a way to do it without the 'for' loop?

% Here is my attempt to get NaN to overwrite all elements along the corresponding FIRST dimension
C = isnan(B); % create a logical mask
D = any (C,1); % look for any NaNs along the first dimension
for k=1:(size(B,1))
    E=squeeze(B(k,:,:));
    E(squeeze(D))=NaN; % apply the 'any' mask
    B(k,:,:)=E ;
end

There is a very similar question here: https://uk.mathworks.com/matlabcentral/answers/358514-replace-all-the-array-with-nan-if-any-of-the-value-is-nan but I couldn't successfully tailor Jan's solution to my question. Thank you.

Answer 1

[Stephen23]

"I wondered if there was any neat trick to do it by array multipication..."

A = 9* ones (6,4,3); % create a 3 dimensional array
A(4,3,2) = NaN; % add a NaN
A(3,2,1) = NaN; % add another NaN;
B = A

B = 
B(:,:,1) =

     9     9     9     9
     9     9     9     9
     9   NaN     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9


B(:,:,2) =

     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9   NaN     9
     9     9     9     9
     9     9     9     9


B(:,:,3) =

     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9

X = 0./~any(isnan(B),1);
B = B+X

B = 
B(:,:,1) =

     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9


B(:,:,2) =

     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9


B(:,:,3) =

     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9

If you really want to use multilplication add 1 to X.

[Stephen23]

A = 9*ones(4,3,6); % create a 3 dimensional array
A(3,2,4) = NaN; % add a NaN
A(2,1,3) = NaN % add another NaN;

A = 
A(:,:,1) =
     9     9     9
     9     9     9
     9     9     9
     9     9     9
A(:,:,2) =
     9     9     9
     9     9     9
     9     9     9
     9     9     9
A(:,:,3) =
     9     9     9
   NaN     9     9
     9     9     9
     9     9     9
A(:,:,4) =
     9     9     9
     9     9     9
     9   NaN     9
     9     9     9
A(:,:,5) =
     9     9     9
     9     9     9
     9     9     9
     9     9     9
A(:,:,6) =
     9     9     9
     9     9     9
     9     9     9
     9     9     9

B = A;
X = 0./~any(isnan(B),3);
B = B+X %            ^ you just need to change this

B = 
B(:,:,1) =
     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9
B(:,:,2) =
     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9
B(:,:,3) =
     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9
B(:,:,4) =
     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9
B(:,:,5) =
     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9
B(:,:,6) =
     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9

Answer 2

[Voss]

You can replicate your matrix D=any(isnan(B),1) in the first dimension using repmat, generating a 3D logical array the same size as B, and then use that 3D logical array as a logical index in B:

A = 9* ones (6,4,3); % create a 3 dimensional array
A(4,3,2)=NaN; % add a NaN
A(3,2,1)=NaN; % add another NaN;
B = A;

B(repmat(any(isnan(B),1),[size(B,1) 1 1])) = NaN;

disp(B);

(:,:,1) =

     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9
     9   NaN     9     9


(:,:,2) =

     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9
     9     9   NaN     9


(:,:,3) =

     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9
     9     9     9     9

[Steve Francis]

Thanks very much for your quick response. I wondered if there was any neat trick to do it by array multipication but your way works well.

[Steve Francis]

I've changed the orientation of the matrix and was hoping that you could advise on the revised repmat statement, please?

A = 9* ones (4,3,6); % create a 3 dimensional array
A(3,2,4)=NaN; % add a NaN
A(2,1,3)=NaN; % add another NaN;
B = A;

The desired output is now

val(:,:,1) =

     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9


val(:,:,2) =

     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9


val(:,:,3) =

     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9


val(:,:,4) =

     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9


val(:,:,5) =

     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9


val(:,:,6) =

     9     9     9
   NaN     9     9
     9   NaN     9
     9     9     9

The following seems to work:

B(repmat(any(isnan(B),3),[1 1 size(B,3)])) = NaN;

Is that correct?