KMeans does not work with numpy.ndarray

[sgerbe]

• scikit-network version: scikit-network==0.28.1
• Python version: 3.9
• Operating System: Windows

Description

Trying to use sknetwork.clustering.KMeans() with an numpy.ndarray gives me a ValueError: Different number of diagonals and offsets.

What I Did

kmeans = KMeans(n_clusters=nclust)
labels = kmeans.fit_transform(adjacency) # adjacency is type numpy.ndarray

This fixes the problem:

adjacency = sparse.csr_matrix(adjacency)
kmeans = KMeans(n_clusters=nclust)
labels = kmeans.fit_transform(adjacency) # adjacency is type numpy.ndarray

I am guessing KMeans.fit() does not assign the check_format back to the input matrix:

Is (Line 113):

self._init_vars()

# input
check_format(input_matrix)

Maybe should be:

self._init_vars()

# input
input_matrix = check_format(input_matrix)

[tbonald]

Thanks for the notice! This has be fixed on the Develop branch.