Correlation nans

Hi all,

I have a very large matrix of 1's, -1's and 0's. When I do corrcoef sometimes I get nans. Is there a way I can avoid getting nans?

Answer 1

[Oleg Komarov]

0/0
ans =
   NaN

Since corr(A, B) = cov(A,B)/(std(A) *std(B)) when you have that the two segments are all zeros, then you will get NaN.

You cannot avoid it but you can decide that the NaN is informative in that case.