Zscore a matrix with NaN

I have matrix A, and I need to find a zscore value (with the intention of running a PCA) of a matrix that has NaN values. I need to have matlab ignore the NaNs while solving for the zscore value. How can I do this? I tried Zscore = zscore(~isnan(A)) but it didn't work.

Thanks for your help!

Answer 1

[Star Strider]

You can create your own function to compute the Z-scores omitting NaN values (I believe the 'omitnan' option began in R2014b). The result vector retains the NaN values while it computes the Z-score without them:

zscor_xnan = @(x) bsxfun(@rdivide, bsxfun(@minus, x, mean(x,'omitnan')), std(x, 'omitnan'));

q = rand(10,2);
q([3 5 7],1) = NaN;
Zscore = zscor_xnan(q);

[Patrick]

can you perhaps elaborate a little bit. I'm new but here's what I had for the function (using MatLab 2015)

 function zcor_xnan = @(x) bsxfun(@rdivide, bsxfun(@minus, x, mean(x,'omitnan')), std(x, 'omitnan'));
end

Here is what I have in my script

if true
  % code
q = rand(10,2);
q([3 5 7],1) = NaN;
Zscore = zscor_xnan(q);
end

Really sorry I know this is a very infantile question but I rarely use functions.

Answer 2

[jkr]

A simple approach for a vector 'x':

zscore = (x - nanmean(x))/nanstd(x);