Hi, The supercomputer admin guys got back to me today and we worked out what the problem was. They have a 'module' system where you can load various different libraries (various different versions of the GCC and Intel compiler suites, and various other tools and libraries) and it was configured so that it loaded some strange libraries by default with python, but without loading *all* of the required libraries. Basically, by forcing it to just load a normal GCC compiler, I could reinstall (which recompiled using GCC) and then it worked fine. Thanks, Robin On Tuesday, 20 October 2015 00:17:24 UTC+1, Josh Warner wrote:

My first question is this: Have you reinstalled or recompiled since the system was upgraded? This kind of error can crop up when you change compilers or the underlying system significantly enough that the previous compilation is no longer understood.

If not, that's the first thing to try as it's simple and easy.

See also these links for more info about that error and troubleshooting compilers and environments:

https://software.intel.com/en-us/forums/intel-fortran-compiler-for-linux-and... https://mail.scipy.org/pipermail/numpy-discussion/2011-August/058089.html

Let us know how it goes!

Josh

On Sunday, October 18, 2015 at 8:09:59 PM UTC-5, Robin Wilson wrote:

...

Hi,

I'm trying to get skimage working on my university supercomputer. I'm sure I've had a previous version working - but unfortunately the supercomputer has been upgraded since then. I've installed via pip, but get the following error when I import threshold_otsu from skimage.filters:

ImportError: /home/rtw1v07/.local/lib/python2.7/site-packages/skimage/filters/rank/generic_cy.so: undefined symbol: _intel_fast_memset

I've reported this to the supercomputer support people, as I suspect it may be some sort of compiler issue - but they're taking ages to respond so I just wondered if anyone here might have any ideas. Does anyone have an idea as to where I should start looking?

Cheers,

Robin