sorting without moving NaNs

How can I sort an array without moving the NaN elements?

I have A=[20 10 NaN 66 NaN 12] and I would like to get A=[66 20 NaN 12 NaN 10]. Thank you for your help. Alessio.

Answer 1

[Jos (10584)]

Use a variant of the same trick:

A=[6 20; 10 10; 3 NaN; 20 66; 4 NaN; 7 12]
B = flipud(sortrows(A,[2 1])) ; % descending sortrows based on 2nd column
A(~isnan(A(:,2)),:) = B(~isnan(B(:,2)),:)

[Image Analyst]

Sebastiano's "Answer" moved here since it's a comment, not an answer:

Thanks a lot, it works perfectly!

Answer 2

[Jos (10584)]

A=[20 10 NaN 66 NaN 12]

B = sort(A,'descend')
A(~isnan(A)) = B(~isnan(B))

Answer 3

[Azzi Abdelmalek]

A=[20 10 NaN 66 NaN 12] 
idx=~isnan(A); 
B=sort(A,'descend');
B(isnan(B))=[];
A(idx)=B

Answer 4

[Jan]

If A is large and contains a lot of NaN's, excluding them from the sorting can save some time:

A      = [20 10 NaN 66 NaN 12] 
idx    = ~isnan(A);
A(idx) = sort(A(idx), 'descend');

[Jos (10584)]

I doubt that excluding NaNs is faster, Jan. Increasing the proportion of NaNs seems to have little effect, or even a positive effect sometimes:

a = randperm(1e7) ; a(a(1:1e1)) = NaN ; tic ; sort(a) ; toc ;
a = randperm(1e7) ; a(a(1:1e5)) = NaN ; tic ; sort(a) ; toc ;

Answer 5

[Sebastiano delre]

I am sorry. Actually I wrongly state my problem. I have A=[6 20; 10 10; 3 NaN; 20 66; 4 NaN; 7 12] and I would like to get A=[20 66; 6 20; 3 NaN; 7 12; 4 NaN; 10 10]. This means I would like to sort the rows of the matrix in a descending order based on the second column and without moving the rows with the NaN elements in the second column. I would like to use sortrows. Sorry again. Alessio.