Failure of doctest for sklearn.cluster._bicluster.SpectralCoclustering on osx.

[lakrish]

Description

DocTest failure on osx when testing master.

______________________________________________________________________________
___________ [doctest] sklearn.cluster._bicluster.SpectralCoclustering 
______________________________________________________________________________
____________
253         The bicluster label of each column.
254 
255     Examples
256     --------
257     >>> from sklearn.cluster import SpectralCoclustering
258     >>> import numpy as np
259     >>> X = np.array([[1, 1], [2, 1], [1, 0],
260     ...               [4, 7], [3, 5], [3, 6]])
261     >>> clustering = SpectralCoclustering(n_clusters=2, random_state=0).fit(X)
262     >>> clustering.row_labels_
Expected:
    array([0, 1, 1, 0, 0, 0], dtype=int32)
Got:
    array([0, 0, 1, 0, 0, 0], dtype=int32)

Steps/Code to Reproduce

pytest sklearn

Expected Results

Actual Results

Versions

System:
python: 3.7.5 (default, Oct 25 2019, 10:52:18) [Clang 4.0.1 (tags/RELEASE_401/final)]
executable: /anaconda3/envs/sklearndev/bin/python
machine: Darwin-18.7.0-x86_64-i386-64bit

Python deps:
pip: 19.3.1
setuptools: 41.6.0.post20191030
sklearn: 0.22.dev0
numpy: 1.16.2
scipy: 1.3.1
Cython: 0.29.13
pandas: None
matplotlib: 3.1.1
joblib: 0.14.0

[TomDLT]

Cannot reproduce on my machine:

System:
    python: 3.7.5 (default, Oct 25 2019, 15:51:11)  [GCC 7.3.0]
executable: /home/tom/miniconda3/bin/python
   machine: Linux-4.15.0-66-generic-x86_64-with-debian-buster-sid

Python deps:
       pip: 19.3.1
setuptools: 41.6.0.post20191030
   sklearn: 0.22.dev0
     numpy: 1.17.2
     scipy: 1.3.1
    Cython: 0.29.14
    pandas: None
matplotlib: 3.1.1
    joblib: 0.14.0

[lakrish]

Just updated numpy and Cython, still get the error.

System:
    python: 3.7.5 (default, Oct 25 2019, 10:52:18)  [Clang 4.0.1 (tags/RELEASE_401/final)]
executable: /anaconda3/envs/sklearndev/bin/python
   machine: Darwin-18.7.0-x86_64-i386-64bit

Python deps:
       pip: 19.3.1
setuptools: 41.6.0.post20191030
   sklearn: 0.22.dev0
     numpy: 1.17.2
     scipy: 1.3.1
    Cython: 0.29.14
    pandas: None
matplotlib: 3.1.1
    joblib: 0.14.0

Could this be a Clang vs GCC issue? Also, see some strangeness reported with Spectral Clustering in this issue #15439

[ogrisel]

I also observed the same failure randomly happening on the joblib CI on Linux AMD64 with python 3.7.4 and numpy 1.16.5 (from the anaconda channel with MKL), scipy 1.3.3 (from PyPI with openblas).

[thomasjpfan]

This is sometimes failing on our pylatest_conda_mkl instance as well.

[thomasjpfan]

Ah it is already reported in #13762 and its been failing more often recently on PRs.