Re: FW: update on covcorm() function, ticket NCL-1681

From: Mary Haley <haley_at_nyahnyahspammersnyahnyah>
Date: Mon Apr 15 2013 - 08:29:56 MDT

Hi All,

Will is correct: the "iopt" variable in corcovm behaves the exact opposite of what the documentation states.

That is, if you use an iopt of (/0,1/), then it is behaving as if you had entered (/1,0/), and vice versa for iopt=(/1,0/). Thus, you will get the incorrect size *and* results returned.

If you used an iopt of (/0,0/) or (/1,1/), then you are okay.

Sorry for the confusion. This will be fixed in the next release of NCL, but I'm not sure yet if we will fix the documentation to match what the function is currently doing, or fix the function to behave the way the documentation currently states.

If we just fix the documentation, then the work-around will work for V6.1.2 and future versions of NCL, so this seems to be the better choice.

I will make a note in the documentation and the "future release" section of "what's new" about the work-around.


On Apr 3, 2013, at 12:23 AM, Will Hobbs wrote:

> Just an update on this, I think the problem is that the 'iopt' arguments are the inverse of what's stated on the help file for covcorm(), i.e. iopt(0) sets the output matrix shape, and iopt(1) sets whether the covariance/correlation matrix is returned.
> You will want to check this yourself, but when I set iopt = (/0,1/) I get an (brow*(nrow+1)/1) matrix with a max value of 1 (indicating a correlation rather than covariance), and when I set iopt=(/1,0/) I get an (nrow,nrow) matrix with values greater than 1.
> Will
> _______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:

ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
Received on Mon Apr 15 08:30:40 2013

This archive was generated by hypermail 2.1.8 : Mon Apr 15 2013 - 20:12:25 MDT