why is corrcoef returning NaN's?

I'm passing in a single matrix, lots of rows (est 5k), 2 cols -- there are no NaNs in the data, and the data is all doubles, for example:

101.340000000000  139.359000000000
101.500000000000  139.630000000000
101.440000000000  139.920000000000
101.350000000000  139.950000000000
101.135000000000  139.940000000000

any thoughts? thanks

Answer 1

[the cyclist]

I suggest you do one of two things to debug this:

(1)

Upload the actual data here, in a MAT file

(2)

Use corrcoef on sections of your matrix at a time. For example, if your matrix is M, and is 5000x2, then try

corrcoef(M(1:2500,:))

and

corrcoef(M(2501:end,:))

At least one of those two will again give a NaN result. Suppose it is the first one. Then you could similarly split it

corrcoef(M(1:1250,:))

and

corrcoef(M(1251:2500,:))

and so on until you isolate a smaller matrix that give the problem, which you may than be able to just eyeball.

[59morgan]

ah. right. there were some nans in there... thx

[the cyclist]

I guess I should have first suggested

any(isnan(M(:)))

Glad it worked out.