Last non nan observation

Hi,

I have a matrix and I want to find in each column the last non nan observation. Is there a neat way of doing it quickly? Regards

Answer 1

[Jiro Doke]

I'm not sure if you're looking for the indices (row number) or the actual value of the last non nan observation. So here are both:

 B = ~isnan(A);

 % indices
 Indices = arrayfun(@(x) find(B(:, x), 1, 'last'), 1:size(A, 2));

 % values
 Values = arrayfun(@(x,y) A(x,y), Indices, 1:size(A, 2));

[Sven]

This will hit a snag if any column is _all_ NaN (the arrayfun will be returning non-uniform output), but otherwise works great.

[Poulomi]

how to overcome this snag @Sven ?

[Sven]

See my answer below for a file exchange entry

Answer 2

[Sven]

Here I can plug a file exchange entry:

http://www.mathworks.com/matlabcentral/fileexchange/27755-n-dimensional-find

Im = rand(50);
Im(Im>.7) = nan;
lastNans = find_ndim(~isnan(Im),1,'last')

To quote: I = FIND_NDIM(BW,DIM,'last') returns the subscripts of the last nonzero elements of BW along the dimension DIM.

The solution hijacks the max() function... it's only a few lines but it packages up things nicely (such as returning 0 if no NaNs were found in a column) - hope you like it.

And if you don't want to use the function itself you can get your answer with just the relevant lines as follows:

BW = ~isnan(Im);
[~, foundPx] = max(BW,[],1);
foundPx = size(BW,1)+1 - foundPx; % Need to count backwards to get last non-zero
foundPx(~any(BW,dim)) = 0; % Account for all-zero entries by setting output to 0