folks, just to follow up on this in case anyone else runs into the same problem: after much mucking around, i conclude that the issue is due to a change that has taken place in numpy between versions numpy-1.9 and numpy-1.10. in my experience, scikit-learn-0.17 cannot be successfully built against numpy-1.10; neither can versions 0.16, 0.15 nor 0.14 for that matter.
therefore, the solution for me was to use numpy-1.9.2 instead of numpy-1.10.2; once this change was made, not only could scikit-learn-0.17 be built but the described unittest issues disappeared and "nosetests sklearn --exe" was able to run completely and successfully. i'm not sure what this means for numpy & scikit-learn going forward in terms of what/where a fix is needed; but hopefully for the moment, others can be saved the trouble of facing the same struggle that i did last week. sincerest thanks to those who took the trouble to help me sort it out... best regards- joe ------------------------------ *From:* Joe Cammisa <jcamm...@haverford.edu> *Sent:* Sunday, December 27, 2015 5:58 PM *To:* scikit-learn-general@lists.sourceforge.net *Subject:* [Scikit-learn-general] scikit-learn-0.17, atlas/unittest issues... greetings. i'm trying to build scikit-learn-0.17 into a homemade python stack on ubuntu-14.04 and having a terrible time of it. no matter what i try, the unittests invariably fail with such messages as these: ERROR: Failure: ImportError (/usr/local/python/lib/python2.7/site-packages/sklearn/metrics/pairwise_fast.so: undefined symbol: ATL_dasum) ERROR: Failure: ImportError (/usr/local/python/lib/python2.7/site-packages/sklearn/metrics/pairwise_fast.so: undefined symbol: ATL_dcopy) ERROR: Failure: ImportError (/usr/local/python/lib/python2.7/site-packages/sklearn/metrics/pairwise_fast.so: undefined symbol: ATL_dscal)) i have a custom built atlas3.10.0/lapack-3.4.2 against which i've successfully built gobs of other software; i also have the ubuntu packages libatlas-base-dev, libatlas-dev, etc, on hand and "ar -t" tells me that the objects in question are present in both. i'm using python-2.7.11, numpy-1.10.2 and scipy-0.16.1, all of which build without problems and pass all unittests as expected. at this point i'm totally stumped and would so gratefully appreciate any advice anyone could offer. warm thanks in advance and happy holidays to all. -joe
------------------------------------------------------------------------------
_______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
------------------------------------------------------------------------------
_______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
