Replace rows with NaN only if there are more than two continous zero values in the same column

I am a begginer in Matlab. I need to replace with NaN all rows which contain more than two continous zero values in the same column (second column). Look at the next example:

Input=

124.2  0  7.2  -4.8
131.1  1.8  1.4  -1.2
131.9  0  1  3
123  0  4  5
987  0  0  9
2323  3  3  3
2323  0  3  2
3  0  6  0
221  3.1  4  5.6
57  1  231  122
987  0  2  7
4454  0  3  4
3  0  0  2
434  0  2  0

Output=

124.2  0  7.2  -4.8
131.1  1.8  1.4  -1.2
131.9  NaN  1  3
123  NaN  4  5
987  NaN  0  9
2323  3  3  3
2323  0  3  2
3  0  6  0
221  3.1  4  5.6
57  1  231  122
987  NaN  2  7
4454  NaN  3  4
3  NaN  0  2
434  NaN  2  0

Thanks for your help fellows!

Answer 1

[Andrei Bobrov]

t = Input(:,2) == 0;
[ii,c] = bwlabel(t);
x = accumarray(ii+1,1);
x = x(2:end);
z = 1:c;
Input(ismember(ii,z(x > 2)),2) = nan;

for all Input

t = Input == 0;
z = cumsum(diff([zeros(1,size(Input,2));t]) == 1);
ii = bsxfun(@plus,z,cumsum([0,z(end,1:end-1)])).*t;
b = accumarray(ii(:)+1,1);
Output = Input;
Output(ismember(ii,find(b(2:end)>2 ))) = nan;

Answer 2

[KSSV]

A = [124.2  0  7.2  -4.8
131.1  1.8  1.4  -1.2
131.9  0  1  3
123  0  4  5
987  0  0  9
2323  3  3  3
2323  0  3  2
3  0  6  0
221  3.1  4  5.6
57  1  231  122
987  0  2  7
4454  0  3  4
3  0  0  2
434  0  2  0];

c2 = A(:,2) ;   % pick second column 

temp = diff(c2 == 0 ) ;
bs = find( temp == 1 ) + 1 ;   % start of zero 
be   = find( temp == -1 ) ;    % end of zero 
be = [be(2:end) ; length(c2)] ;
%%replace with Nan's
for i = 1:length(bs)
    c2(bs(i):be(i)) = NaN ;
end
A(:,2) = c2 ;

Answer 3

[Gareth Lee]

First, you can find the column with continous zeros(more than two), then find the index for replacement with nan. for example:

a =(A==0); 
a = double(a); % you can loop to find the number of 1 (more than 2)
m = a(:,2);
[cc,dd] = regexp(sprintf('%d', m), '1{3,}', 'start', 'end'); % find the index of continous ones

Next, replace the zeros between cc and dd with nan.

[Jan]

The conversion between DOUBLEs and CHARs is time consuming and prone to errors and rounding for floating point values. Better stay at the same type.

Answer 4

[Walter Roberson]

http://www.mathworks.com/matlabcentral/answers/307780-replace-with-the-mean-if-there-are-at-least-six-continous-nan-in-the-same-column

Odd how two different people have such similar tasks for the same data...

Answer 5

[Jan]

With https://www.mathworks.com/matlabcentral/fileexchange/41813-runlength:

[B, N]             = RunLength(Data(:, 2));
B(B == 0 & N >= 3) = NaN;
Data(:, 2)         = reshape(RunLength(B, N), [], 1);