How to Normalize a matrix between 0-1 containing NaNs?

Can anyone help me please to normalize a matrix between 0-1 which contains NaNs?

the following makes whole the matrix zero

d =  A - min (A(:));
nrmdata = d./max(d(:));

[Walter Roberson]

I do not seem to be able to reproduce the problem. Could you attach a .mat with your A values?

[ML]

the matrix is attached, it contains both inf and NaN.

Answer 1

[KSSV]

    clc; clear all ;
    load A.mat ;
    idx = isinf(A) ;  % get the positions of inf in A
    B = A ;   % B as A 
    B(idx) = 0 ;  % repalce inf's in B with 0 
    % normalize the B matrix 
    norm_A = (B - min(B(:))) / ( max(B(:)) - min(B(:)) ) ;
    norm_A(idx) = inf ;   % replace norm_A with inf in previous locations/ if  required

or

clc; clear all ;
load A.mat ;
m0 = min(A(isfinite(A(:))));   % get minimum in A
m1 = max(A(isfinite(A(:))));   % get maximum in A
norm_A = (A -m0)/(m1-m0 ) ;

Answer 2

[Walter Roberson]

You cannot normalize data that contains infinity, not unless you are willing to ignore the infinity.

By definition, any finite real value is a negligible fraction of infinity, so when you normalize against infinity, all finite values become 0.

Answer 3

[Stephen23]

Ignoring infinity:

S = load('A.mat');
A = S.A;
idx = isfinite(A);
Amax = max(A(idx));
Amin = min(A(idx));
B = (A-Amin) / (Amax-Amin);

and checking:

>> min(B(isfinite(B)))
ans = 0
>> max(B(isfinite(B)))
ans =  1