Difference between IRAF from STSci and NOAO? On OS X?
hawcheng wrote on Mar 22, 2010
I have just installed IRAF for OS X, which I downloaded from STSci. I have found that the results I get using this on my personal computer are not exactly the same as when I use IRAF on my department's Linux machines, which I understand is the standard NOAO installation (version 2.14.)
Specifically, I find that it is often the case that when I use routines such as zerocombine, flatcombine, and ccdproc, in either their noao.imred.ccdred or mscred incarnations, that the final pixel values will differ by as much as 0.007 ADU. The average discrepancy is on the order of 10^-6 or 10^-8, but the fact that it occasionally strays into the thousandths concerns me.
Has anyone else seen this or can comment on this?
Specifically, I find that it is often the case that when I use routines such as zerocombine, flatcombine, and ccdproc, in either their noao.imred.ccdred or mscred incarnations, that the final pixel values will differ by as much as 0.007 ADU. The average discrepancy is on the order of 10^-6 or 10^-8, but the fact that it occasionally strays into the thousandths concerns me.
Has anyone else seen this or can comment on this?
Mike Fitzpatrick wrote on Mar 22, 2010
You'd have to check with STScI (help@stsci.edu) whether they recompiled the release or made other changes (I doubt it, though).
The difference may be as simple as v2.14 vs v2.14.1 which should be visible on the banner when you first log in, or you can check the dates on the binaries (from Sep08 it's v2.14.1, earlier an it's v2.14).
That said, it is not unexpected that there are floating point differences between OSX and Linux systems, it sounds like the differences are within the floating point precision and not some gross error so I'm not sure there's a problem.
-Mike
The difference may be as simple as v2.14 vs v2.14.1 which should be visible on the banner when you first log in, or you can check the dates on the binaries (from Sep08 it's v2.14.1, earlier an it's v2.14).
That said, it is not unexpected that there are floating point differences between OSX and Linux systems, it sounds like the differences are within the floating point precision and not some gross error so I'm not sure there's a problem.
-Mike
hawcheng wrote on Mar 22, 2010
All right, I'll check with them.
In the meantime, I've noticed that when I use perform bias subtraction and flatfielding "manually" with mscarith or imarith, that the results between what I get using the OS X STScI version and the Linux version come out the same, unlike what happens when I use ccdproc. Any ideas on what this might mean?
In the meantime, I've noticed that when I use perform bias subtraction and flatfielding "manually" with mscarith or imarith, that the results between what I get using the OS X STScI version and the Linux version come out the same, unlike what happens when I use ccdproc. Any ideas on what this might mean?
Mike Fitzpatrick wrote on Mar 22, 2010
CCDPROC does the bias based on the overscan region, IMARITH et al do a pixel-for-pixel subtraction of the entire image.
hawcheng wrote on Mar 22, 2010
Ah, yes. That I understand. My point was more that if it were in fact some kind of floating point difference between the OSX and Linux operating systems, wouldn't I expect to see that when I'm using imarith or mscarith as well, just like when I'm using ccdproc?
Also, I did check the versions and they both appear to be IRAF 2.14.
Also, I did check the versions and they both appear to be IRAF 2.14.
Last post on Mar 22, 2010