Replace NaNs with next real value.

How can I replace all NaN values in an array with the next non-NaN value in the array? I have

x = [NaN 1 1 3 NaN NaN 4];

and I'd like to turn x into

x = [1 1 1 3 4 4 4];

The following works, but I'd like to do it with fewer moving parts:

for k = find(isnan(x))
    x(k) = x(find(1:length(x)>k & isfinite(x),1,'first'));
end

Answer 1

[Chad Greene]

Thanks for your suggestions, y'all. It's nice that R2014b has the 'next' and 'previous' options for 1D interpolation. I ended up writing a function to do the job in old 2012b.

Answer 2

[Image Analyst]

I wouldn't worry about it. If it works, that way is fine. If you have millions of elements or a 2D image, then there are functions that can do it in the Image Processing Toolbox, like imfill().

Answer 3

[Sean de Wolski]

x = [NaN 1 1 3 NaN NaN 4]; 

idx = isnan(x);
idxnan = find(idx);
idxnotnan = find(~idx);
fillv = interp1([-flintmax idxnotnan],[0 x(idxnotnan)],idxnan,'next');
x(idx) = fillv;

Probably slower and harder to figure out than a for-loop and it requires R2014b; but it's cool!

[yogan Sganzerla]

Where was the variable flintmax defined?

[Image Analyst]

@yogan Sganzerla, flintmax is built-in:

>> flintmax

ans =

9.00719925474099e+15

Answer 4

[per isakson]

If &nbsp x(end) &nbsp is a number

    while any( isnan( x ) )
        x1 = [ x(2:end), inf ];
        x( isnan(x) ) = x1( isnan(x) ); 
    end

or

    while any( isnan( x ) )
        isn = isnan(x);
        x( isn ) = x( [false,isn(1:end-1)] ); 
    end

[yogan Sganzerla]

However, if the vector is like the following, it doesn't work properly.

x = [NaN 1 1 3 NaN NaN 4 NaN NaN NaN];

Where the output will be:

[1 1 1 3 4 4 4 NaN NaN NaN]

and my intentions is to be:

[1 1 1 3 4 4 4 4 4 4].

So repeating the last real value until the end (if the vector just has NaN).

What should I add to reach this output?

Thank you for your attention.

[Image Analyst]

@yogan Sganzerla, try this:

x = [NaN 1 1 3 NaN NaN 4 NaN NaN NaN]
nanIndexes = find(isnan(x))
nonNanIndexes = find(~isnan(x))
% Make any leading nans the first non nan value
if nonNanIndexes(1) ~= 1
	x(1:nonNanIndexes(1) - 1) = x(nonNanIndexes(1))
end
% Make any trailing nans the last non nan value
if nonNanIndexes(end) ~= length(x)
	x(nonNanIndexes(end) + 1 : end) = x(nonNanIndexes(end))
end
% Now regenerate indexes on the changes vector
nanIndexes = find(isnan(x))
nonNanIndexes = find(~isnan(x))
% Replace nan indexes with last non-nan value just prior to it.
for k = 1 : length(nanIndexes)
	thisIndex = nanIndexes(k);
	x(thisIndex) = x(thisIndex - 1)
end

Answer 5

[Roger Stafford]

Here's another way:

   t = isnan(x);
   f1 = find(t);
   f2 = find(diff([t,false])==-1)+1;
   f3 = zeros(1,length(f1));
   f3(find(diff([-1,f1])>1)) = diff([0,f2]);
   f3 = cumsum(f3);
   x2 = [x,x(end)];
   x(f1) = x2(f3);