remove first s and last t rows of a matrix containing NaN, leave lows in the middle containing NaN.

I have a Tx2 Matrix A and I would like to remove the rows in the beginning and in the end that contain any NaN. For example:

A=[[NaN;NaN;NaN;4;1;NaN;5;6;8;NaN;NaN],[NaN;NaN;2;7;6;5;NaN;6;18;2;NaN]]

should the equal to:

A=[[4;1;NaN;5;6;8],[7;6;5;NaN;6;18]]

many thanks for your help, Jo.

Answer 1

[Thorsten]

There might be smarter solutions to figure out the indices of leading and trailing 1's in nanflag, but this solution works:

A=[[NaN;NaN;NaN;4;1;NaN;5;6;8;NaN;NaN],[NaN;NaN;2;7;6;5;NaN;6;18;2;NaN]];
nanflag = isnan(sum(A'));
ind = []; for i = 1:numel(nanflag), if nanflag(i) == 1, ind = [ind i]; else break, end, end
for i = numel(nanflag):-1:1, if nanflag(i) == 1, ind = [ind i]; else break, end, end

Anew = A(setdiff(1:size(A,1), ind), :);

[Jo]

many thanks, this code works well

Answer 2

[Guillaume]

A=[[NaN;NaN;NaN;4;1;NaN;5;6;8;NaN;NaN],[NaN;NaN;2;7;6;5;NaN;6;18;2;NaN]];
removestart = logical(sum(cumprod(isnan(A)), 2));
removeend = flipud(logical(sum(cumprod(flipud(isnan(A))), 2)));
A(removestart | removeend, :) = []

[Jo]

When I enter this code I get the Error: "Error using cumprod CUMPROD is only supported for floating point input."

[Guillaume]

Works fine on R2014b, which version are you using?

In any case if cumprod does not accept logical, just convert them to double:

 A=[[NaN;NaN;NaN;4;1;NaN;5;6;8;NaN;NaN],[NaN;NaN;2;7;6;5;NaN;6;18;2;NaN]];
 nanpos = double(isnan(A));
 removestart = logical(sum(cumprod(nanpos), 2));
 removeend = logical(sum(cumprod(nanpos, 'reverse'), 2));
 A(removestart | removeend, :) = []

I've also simplify the calculation of removeend. I didn't realise that cumprod had a reverse option.

Shame you accepted a less efficient code.

[Jo]

I am using R2012b, this may be the cause. For anyone having the same version, adding double(..) helps, this adapted code works for me:

   A=[[NaN;NaN;NaN;4;1;NaN;5;6;8;NaN;NaN],[NaN;NaN;2;7;6;5;NaN;6;18;2;NaN]];
   nanpos = double(isnan(A));
   removestart = logical(sum(cumprod(nanpos), 2));
   removeend = flipud(logical(sum(cumprod(flipud(nanpos)), 2)));
   A(removestart | removeend, :) = []

Thanks a lot!

Answer 3

[C.J. Harris]

Nobody should ever need more than one line:

A  = [[NaN;NaN;NaN;4;1;NaN;5;6;8;NaN;NaN],[NaN;NaN;2;7;6;5;NaN;6;18;2;NaN]];

A2 = A(find(~any(isnan(A),2),1,'first'):find(~any(isnan(A),2),1,'last'),:);