general-discussion > Different results - niak_test_all
Showing 1-2 of 2 posts
Display:
Results per page:
May 18, 2016  05:05 PM | Raphael Casseb
Different results - niak_test_all
Dear experts,

I am new to NIAK and I have just run the niak_test_all.
I am running it on docker in a Mac OS X (10.9.5), GNU Octave 3.8.1
There are many 0 results. How do I know if I should worry about it? I know that I should expect different results if I am running in a different configuration, but on NIAK page it says "Note that tests can fail if slightly different versions of Matlab/Octave/minc tools are used when generating results as compared to the reference results. When NIAK is executed through docker, perfect replications are expected."

Any lights?
Thanks in advance,
Raphael
May 18, 2016  06:05 PM | Pierre Bellec
RE: Different results - niak_test_all
Hi Raphael,

I would first like to make sure I understand what you have done. Did you run the test through docker? In that case the octave that is used is whatever ships with niak, not the one you may have installed on mac OSX. Would you mind copying the line you are using to invoke niak in docker? that will help me understand what version of the niak image you used (that includes the OS, octave, the minc tools etc, not just the niak library per say).

Also, could you please attach a copy of the test report that includes differences. In my experience so far, docker has lead to perfect replications across systems. But it is not impossible that some differences will creep up across systems even when running niak inside a container, I just haven't observed it yet. The report contains statistics on the magnitude of the differences, that will help me give you feedback on the severity of the differences that are occuring. 

Best,

Pierre