Running standard deviation on matrix with NaN values

Hello, I have large matrix where each row represents time series for one location. I need to get equally sized matrix containing running standard deviation along the row dimension. There are several function to do this but none allows for NAN values. My time series includes a large amount of gaps and I can not interpolate the data. I would appreciate any suggestions. Thank you

Answer 1

[Jos (10584)]

function SD = nanstdrow(X)
% NANSTDROW - SD per row ignoring NaNs
tf = ~isnan(X) ; % non-nan values
X(~tf) = 0 ; % set NaN values to zero so they do not contribute to the mean
N = sum(tf,2) ; % number of elements per row
M = sum(X,2) ./ N ; % average of row
D = (X - repmat(M,1,size(X,2))).^2 ; % squared difference with mean
SS = sum(tf .* D,2) % row sum 
SD = sqrt(SS./N) ; % calculate SD per row

[Jos (10584)]

not thoroughly tested but it should give you an idea ..

[Walter Roberson]

This uses the biased form of std rather than the unbiased form, which would divide SS by (N-1) instead of by (N)

[Jos (10584)]

True! Use SD = sqrt(SS./(N-1))

[Tereza Smejkalova]

Thanks, in the end I solved it by using nlfilter and static function.

Answer 2

[Walter Roberson]

function s = nanstd(X)
  s = std(X(~isnan(X)));
end

Answer 3

[Tereza Smejkalova]

I know about the nanstd() but if I understand correctly I would have to run in in a loop within loop for each window and each row. I forgot to mention before that my table is 380 000 rows and 1000 columns. I was wondering if there is anything to use without a loop?