coclust.clustering.SphericalKmeans returns AttributeError: 'numpy.ndarray' object has no attribute 'todense'

[ahgraber]

I have a 2D numpy array 'doc_vecs' of document vectors (axis 0 is a document; axis 1 is the vector representation). When I try to pass this to SphericalKmeans I get the following error:

---

AttributeError Traceback (most recent call last)
.py in
7 start = timeit.default_timer()
8 skm_model = SphericalKmeans(n_clusters=k, weighting=False)
----> 9 skm_model.fit(doc_vecs)
10 skm_labels.append(skm_model.labels_)
11 skm_criterion.append(skm_model.criterion)

~/miniconda3/envs/nlp_tm/lib/python3.7/site-packages/coclust/clustering/spherical_kmeans.py in fit(self, X, y)
92 X = transformer.fit_transform(X)
93
---> 94 X = X.todense()
95 X = np.array(X)
96 X = sp.lil_matrix(X)

AttributeError: 'numpy.ndarray' object has no attribute 'todense'

Relevant versioning info:
MacOS 10.14.6
Anaconda Python 3.7.3
numpy = 1.16.4
scipy = 1.3.1
scikit-learn = 0.21.2
matplotlib = 3.1.1
coclust = 0.2.1

Thanks for your help!

[smorbieu]

Thank you for the bug report.

As a workaround until we fix it, you can transform the array into a sparse matrix :

from scipy.sparse.csc import csc_matrix

doc_vecs = csc_matrix(doc_vecs)