Any suggestion to replace the diagonal values of matrices in a 3D array?

Any suggestion to replace the diagonal values of matrices in a 3D array?

in case1 to "0"

in case1 to "NaN"

e.g. for NaN

M_3D=randi(100, 4,4,3); %fake input
N=size(M_3D,1);
mask=1:N+1:end)=nan;
M_3D=M_3D.*mask;

[Steven Lord]

What do you mean by "diagonal" in this context? If I had:

A = reshape(1:27, [3 3 3])

A = 
A(:,:,1) =

     1     4     7
     2     5     8
     3     6     9


A(:,:,2) =

    10    13    16
    11    14    17
    12    15    18


A(:,:,3) =

    19    22    25
    20    23    26
    21    24    27

Do you want to set elements 1, 5, 9, 10, 14, 18, 19, 23, and 27 to 0? This would be setting the elements along the diagonal of each page of the 3-d array to 0.

Or do you just want to set elements 1, 14, and 27 to 0? This would be setting elements along a line "connecting opposite corners" (or the equivalent if your array isn't a cube) to 0.

[Stephen23]

@Steven Lord: out of curiosity, how would you do a "connecting opposite corners" for an arbitrary number of dimensions?

[Steven Lord]

One possibility is points at subscripts n*ones(1, ndimsOfArray) for n = 1:min(size(theArray)). Once you reach any "edge" of the array you stop.

So for an array of size (4, 3, 5) those would be the elements at subscripts (1, 1, 1), (2, 2, 2), and (3, 3, 3). That matches conceptually the behavior of the diag function on non-square matrices.

A = reshape(1:12, 3, 4)

A = 3×4

1 4 7 10 2 5 8 11 3 6 9 12

diag(A) % (1, 1), (2, 2), and (3, 3) since min(size(A)) is 3.

ans = 3×1

1 5 9

[julian gaviria]

@Steven Lord, following you example, I want to replace the elements 1, 5, 9, 10, 14, 18, 19, 23, and 27 either by "0" or by "NaN" values. These are two different cases. Apologies for not providing all the info in the initial post. Sometimes my 3D matrices have NaN values in the diagonals, which I must replace by "0" value.s @Dyuman Joshi's suggestion worked out.

Thanks anyway.

Answer 1

[Dyuman Joshi]

Replacing diagonal values of each 2D matrix of a 3D matrix with NaN

M = randi(100, 4,4,3); %fake input
N = eye(size(M,[1 2]));

M(M&N)=NaN

M = 
M(:,:,1) =

   NaN    88    80    30
    51   NaN    98    73
    75   100   NaN    63
    68    51    61   NaN


M(:,:,2) =

   NaN    13    65    60
    99   NaN    55    50
    16    24   NaN    94
    46    52    86   NaN


M(:,:,3) =

   NaN    50    49    17
    41   NaN    68    52
    79    80   NaN    79
    83    47    86   NaN

[julian gaviria]

Thank you @Dyuman Joshi,

Do you know any alternative when the diagonal values are NaNs? Your code did not work in that situation:

Error using &

NaN values cannot be converted to logicals.

[Dyuman Joshi]

In that case -

%fake input
M = randi(100, 4,4,3);
M(1) = NaN

M = 
M(:,:,1) =

   NaN     7    27    89
    63    89    61    21
    59    79    68    44
    66    45    17    61


M(:,:,2) =

    99    92     2    33
    55    25    24    31
    21    46    88    39
    75    79    12    39


M(:,:,3) =

    47     6    69    13
    70    30    51    49
     6    44    85    63
    23    27    97    91

N = eye(size(M,[1 2]));

M(ones(size(M))&N)=NaN

M = 
M(:,:,1) =

   NaN     7    27    89
    63   NaN    61    21
    59    79   NaN    44
    66    45    17   NaN


M(:,:,2) =

   NaN    92     2    33
    55   NaN    24    31
    21    46   NaN    39
    75    79    12   NaN


M(:,:,3) =

   NaN     6    69    13
    70   NaN    51    49
     6    44   NaN    63
    23    27    97   NaN

[julian gaviria]

It nicely worked! Thanks a lot @Dyuman Joshi

[Dyuman Joshi]

You are welcome!