ImportError: cannot import name '_k_means'

[Randool]

Hi. I install spherecluster usingpip install spherecluster successfully in Ubuntu 18.04. But when I call from spherecluster import SphericalKMeans, I got an ImportError.

Traceback (most recent call last):
  File "LM/vectors_cluster.py", line 9, in <module>
    from spherecluster import SphericalKMeans
  File "/path_to_anaconda/lib/python3.6/site-packages/spherecluster/__init__.py", line 2, in <module>
    from .spherical_kmeans import SphericalKMeans
  File "/path_to_anaconda/lib/python3.6/site-packages/spherecluster/spherical_kmeans.py", line 16, in <module>
    from sklearn.cluster import _k_means
ImportError: cannot import name '_k_means'

Here is my environmental information:

Package	Version
numpy	1.14.3
scipy	1.1.0
scikit-learn	0.22.2.post1
pytest	3.5.1
nose	1.3.7
joblib	0.14.1
spherecluster	0.1.7

If anyone can help me, I would really appreciate it!

[rtrad89]

I'm not 100% sure, but it appears that sklearn 0.22.1 has introduced some changes that made the aforesaid code line to fail, namely:

from sklearn.cluster import _k_means

I got the code running using:

from sklearn.cluster import _kmeans as _k_means  # _k_means directly caused the error

But I am getting some new errors in the function _check_sample_weight, more precisely at the line:

n_samples = _num_samples(X)

The error reads:

TypeError: Expected sequence or array-like, got <class 'NoneType'>

Despite that, I hope this is of any help.

[Randool]

@rtrad89 Thanks. I also made the same changes to the code and got the same 'TypeError'. Hope for the next version.

[rtrad89]

@rtrad89 Thanks. I also made the same changes to the code and got the same 'TypeError'. Hope for the next version.

@Randool would you please confirm whether you have the latest spherecluster version, 0.1.7 that is? Because I noticed something may be the culprit.

In my local version of the aforementioned spherecluster (0.1.7), the call in spherical_kmeans.py is:

sample_weight = _check_sample_weight(X, sample_weight)

While in the online github version, it is in the correct form:

sample_weight = _check_sample_weight(sample_weight, X)

[Randool]

@rtrad89 I read the source code and found this change, so I commented this line. But I remember the result is still disappointing.

[YBZh]

It seems that downgrading the sklearn to the older version of 0.20.0 can solve the problem.
conda install scikit-learn=0.20.0

[Randool]

It seems that downgrading the sklearn to the older version of 0.20.0 can solve the problem.
conda install scikit-learn=0.20.0

Thanks! That's the right solution!

[glicerico]

This is a workaround, not a solution.
A side-effect of downgrading to 0.20.0 is that the OPTICS clustering algorithm is not available anymore.
Can we re-open this issue, and find a proper solution for it?

[glicerico]

Oh, I see this has been taken care of in the develop branch of the original repo, as @rtrad89 pointed out above.
So the next package should solve this issue, and for now one has to compile the repo code to have it working properly.
I tested with OPTICS in the develop version and it works fine :)
@Randool, sorry, I think this issue can indeed be closed.

[Randool]

haha, looking forward to the next version :> @glicerico

[rtrad89]

This is a workaround, not a solution.
A side-effect of downgrading to 0.20.0 is that the OPTICS clustering algorithm is not available anymore.
Can we re-open this issue, and find a proper solution for it?

Just reporting that downgrading only to 0.22.0 doesn't have this side effect.

If you are using Anaconda3, you can write:

 conda install scikit-learn=0.22.0