Replacing variables in a dataset

Hi, I have a dataset e.g like this 44 44 0; 3 3 0; 132 133 0; 45 46 42; and want to replace the zeros with the average of each row (each row represents one variable). I know how to do it for each row individually, was wondering if there is a faster way though for when dealing with a large matrix?? Thanks !

Answer 1

[Guillaume]

If the 0s are supposed to indicate missing values, then you should be using NaN instead. Assuming that the mean is supposed to be taken without the 0s:

m = [44 44 0; 3 3 0; 132 133 0; 45 46 42;]
m(m == 0) = NaN;  %replace 0 with NaN
mu = repmat(mean(m, 2, 'omitnan'), 1, size(m, 2));  %calculate mean and replicate to same size as M
m(isnan(m)) = mu(isnan(m))  %replace NaNs by mean

[Yiannis Raptodimos]

Thank you very much!

Answer 2

[Hoang Nguyen]

Hi,

It seems you are trying to replace all the 0's in a matrix with the mean of the corresponding row. For example:

    44  44  0
    3   3   0
    132 133 0
    45  46  42

becomes

    44   44   29.33
    3    3    2
    132  133  88.33
    45   46   42

This can be done with simple matrix operations:

function[result] = replace0sWithMean(M)
    % Finds the mean column vector by averaging each row
    mu = mean(M, 2); 

    % Adds the mean column vector to the original matrix
    added = M + mu; 

    % added(:,:) ~= mu(:) returns a boolean matrix where the 1's correspond to non-zero's in the original matrix
    % Piece-wise multiply this result by the mean vector
    % Subtract this quantity from the sum computed earlier in order to restore each original non-zero value
    result = added - (added(:,:) ~= mu(:)) .* mu
  end

As Guillaume wonderfully pointed out the case in which we don't want the 0's to be involved in the calculation of the means, use the following code snippet instead:

    function[result] = replace0sWithMean(M)
          temp = M;
          temp(temp==0) = NaN;
          mu = mean(temp, 2, 'omitnan');

          added = M + mu;
          result = added - (added(:,:) ~= mu(:)) .* mu;
     end

[Guillaume]

I would suspect that the mean is to be taken without the 0s.