无法在sklearn中使用稀疏矩阵计算silhouette_score

Question

我尝试使用稀疏矩阵计算silhouette_score或silhouette_samples，但得到以下错误：

ValueError: diag需要至少二维的数组

示例代码如下：

edges = [
(1, 2, 0.9),
(1, 3, 0.7),
(1, 4, 0.1),
(1, 5, 0),
(1, 6, 0),
(2, 3, 0.8),
(2, 4, 0.2),
(2, 5, 0),
(2, 6, 0.3),
(3, 4, 0.3),
(3, 5, 0.2),
(3, 6, 0.25),
(4, 5, 0.8),
(4, 6, 0.6),
(5, 6, 0.9),
(7, 8, 1.0)]

gg = nx.Graph()

for u,v, w in edges:
    gg.add_edge(u, v, weight=w)


adj = nx.adjacency_matrix(gg)
adj.setdiag(0)


from sklearn.metrics import silhouette_score, silhouette_samples

print(silhouette_score(adj, metric='precomputed', labels=labels))
silhouette_samples(adj, metric='precomputed', labels=labels)

Answer 1

这是一个bug。你应该上报。Relevant code.

X, labels = check_X_y(X, labels, accept_sparse=['csc', 'csr'])

# Check for non-zero diagonal entries in precomputed distance matrix
if metric == 'precomputed':
    atol = np.finfo(X.dtype).eps * 100
    if np.any(np.abs(np.diagonal(X)) > atol):
        raise ValueError(
            'The precomputed distance matrix contains non-zero '
            'elements on the diagonal. Use np.fill_diagonal(X, 0).'
        )

尽管输入检查显式接受CSC/CSR矩阵，但如果度量为'precomputed'，它会将X放入不适用于稀疏矩阵的numpy函数中。