1
Vote

Fatal Error : std=.-nan has illegal value for SNPs

description

I am processing dosage data of 7.8 mio imputed SNPs.
Working with the Software version 1.05 i was getting this error message:
Fatal Error : std=.-nan has illegal value for SNPs[:][1085]
 
I couldn't find any unexpected values, format errors or special characters in the input dosage files.
The dosage files were originating from a 64bit unix system.
Because i couldn't find any other solution I excluded ~100 SNPs producing this error.
With those excluded SNPs, the process did run successfully.
 
Now i transfered the successfully running input dosage data together with their RRM and spectral decomposition folders - and on a new server, with FaSTLMM v1.07 i am getting again an error message:
Fatal Error : std=.0.000000 has illegal value for SNPs[:][rs16987431]
 
In the attached output file you can see that the same input data runs on server 1 and does not run on server 2.
Is this a rounding issue of the spectral decompositions which were generated on server 1?
Or a version issue?

file attachments

comments

bobd00 wrote Mar 23, 2012 at 5:14 AM

The v1.05 'std=.-nan' issue came from not eliminating SNPs with no variation (a Single Nucleotide Constant or SNC). This variation of 0.0 may lead to divide by zero problems. This sounds like your v1.05 issues. v1.07 fixed all the issues I knew about for SNCs, but had a significant perf degradation on .bin files.

Regarding the server different outputs, the two outputs you sent in the attached file are from different version of FastLmmC (v1.05 & v1.07). The fact that v1.07 found additional SNCs in your data indicates the ~100 snps you pulled out did not get all of the SNCs and that is why the two are behaving differently.

v1.08 fixes that perf problem, so I'd recommend an upgrade to 1.08 for you and use it to run the experiment again. Please let us know how that works.

wrote Apr 10, 2012 at 5:39 PM

Did the update eliminate your problems here?  If so, let us know and I will close the issue.  If you are still having problems, please let me know that too so I can figure out what is gong on.

costin_leu wrote Apr 11, 2012 at 5:19 PM

The same error still occurs with v1.08 as well as with v1.09.

wrote Feb 13, 2013 at 11:40 PM