Skip to content


Subversion checkout URL

You can clone with
Download ZIP


test_spectral_clustering_sparse fail on OS X 10.8.2 #2472

chyikwei opened this Issue · 12 comments

4 participants


I got this error on sklearn.cluster.tests.test_spectral.test_spectral_clustering_sparse
Any one knows what might be wrong?

OS: Max OS X 10.8.2
numpy: 1.7.1
scipy: 0.11.0

FAIL: sklearn.cluster.tests.test_spectral.test_spectral_clustering_sparse
Traceback (most recent call last):
  File "/Users/chyikwei/getglue-python-webservice/virtual/lib/python2.7/site-packages/nose/", line 197, in runTest
  File "/Users/chyikwei/scikit-learn/sklearn/cluster/tests/", line 155, in test_spectral_clustering_sparse
    assert_greater(np.mean(labels == [1, 1, 1, 1, 1, 0, 0, 0, 0, 0]), .89)
AssertionError: 0.59999999999999998 not greater than 0.89
    """Fail immediately, with the given message."""
>>  raise self.failureException('0.59999999999999998 not greater than 0.89')

I am running a similar setup (except for scipy which is 0.12.0 here) and I cannot reproduce the failure.


I think the issue I met is about "arpack" module.

When I set eigen_solver=None or eigen_solver="arpack" , output from spectral_embedding function looks weird:

[[ -2.32766652e+08 -1.39966513e+00]
[ 3.86210625e+08 -1.54720798e+00]
[ -1.48187517e+08 -9.47395362e-01]
[ -5.64702377e+07 -7.93513447e-01]
[ 1.21781087e+08 -1.27888857e-01]
[ 1.59543207e+08 8.04107066e-01]
[ -2.11367269e+08 9.66087598e-01]
[ -1.47024602e+08 9.62468083e-01]
[ -1.85680793e+06 9.62468083e-01]
[ 1.30138155e+08 1.12053995e+00]]

But when I set eigen_solver="amg" , the output of spectral_embedding looks more reasonable:

[[-3.0003491 -0.76365248]
[-3.31662479 -0.91420026]
[-2.03085492 -0.18276992]
[-1.70099068 0.57040841]
[-0.27414501 3.16227766]
[ 1.72369937 -0.85514476]
[ 2.07092395 1.20605346]
[ 2.06316509 -0.76410269]
[ 2.06316509 -0.76410269]
[ 2.40201099 -0.69476673]]

And the output can be classified correctly by kmeans and pass the test.
Any suggestion how to fix it?


Indeed this is a known bug of OSX that will be fixed in 10.9: See #2431 and scipy/scipy#2547. What is weird is that I don't reproduce the failure myself...


Actually I was mistaken: it has also been fixed in the scipy wrapper itself in the master branch and schedule for release 0.13.


The issue is fixed after I installed scipy ''
But the test still failed at test_connect_regions_with_grid (same as issue #1968)


I could reproduce this issue (with test test_spectral_clustering_sparse) on osx with scipy 0.12.0 but apparently only with python 3, not python 2.7... weird.


I cannot reproduce the test_connect_regions_with_grid failure myself...


Ah ok test_connect_regions_with_grid is only there in scipy master, not in 0.12.0. If you have time to investigate this one your input on #1968 would surely be appreciated.


@ogrisel FYI I can reproduce this with recent Anaconda on Mac OS 1.8 and numpy 1.8 and sicpy 0.12.
Any news on this?


ahhh sorry numpy 1.7.1


@dengemann The test fail is fixed after I re-install scipy 0.12 from its source code.

What I did is

  1. remove scipy
  2. git clone scipy repository
  3. checkout to tag 0.12.0rc1
  4. install

Maybe you can give it a try.

@amueller amueller added this to the 0.15.1 milestone

I cannot reproduce this failure when building scipy against the Accelerate of OSX 10.9.


@ogrisel ogrisel closed this
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.