# Remove NaN from a matrix

8 views (last 30 days)
Damith on 3 Jun 2014
Commented: Damith on 4 Jun 2014
Hi,
I have matrix B = (366 x 39) and has NaN values in every column. How can I remove NaN and obtain the mean of every column. The resulting matrix C should be a 1 x 39 matrix.

George Papazafeiropoulos on 3 Jun 2014
Edited: George Papazafeiropoulos on 3 Jun 2014
ind=sum(~isnan(B));
B(isnan(B))=0;
mean=sum(B)./ind

James Tursa on 3 Jun 2014
If you have the appropriate toolbox, you can use the function: nanmean
Damith on 4 Jun 2014
Thanks

dpb on 3 Jun 2014
If have the Stats Toolbox, then just nanmean
If don't, it's more of a pain; unless the NaN are consistent in number per column you'll end up w/ a "jagged" array which isn't allowed in an array. Hence you'll need to accumulate by column--the "deadahead" solution is a loop and for small matrices of this order it'll undoubtedly be "fast enough"...
nc=size(B,2); % number columns in B
m=zeros(1,nc); % preallocate for the means vector
for i=1:nc
m(i)=mean(isfinite(B(:,i))); % compute mean of values less NaN's and Inf's
end
You can use accumarray for this w/o the loop...seed doc's for details
Damith on 4 Jun 2014
Thanks.

Image Analyst on 3 Jun 2014
You might want to take a look at John D'Errico's inpaint_nans: http://www.mathworks.com/matlabcentral/fileexchange/4551-inpaint-nans