Implementation:Scikit learn Scikit learn SpectralBiclustering

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Biclustering, Unsupervised Learning
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for performing spectral biclustering on data matrices provided by scikit-learn.

Description

SpectralBiclustering partitions rows and columns of a data matrix simultaneously under the assumption that the data has an underlying checkerboard structure. It uses singular value decomposition (SVD) to find the best low-rank approximation of the normalized data matrix, then applies K-Means clustering to the resulting singular vectors. The algorithm supports multiple normalization methods including bistochastic, scale, and log normalization (Kluger, 2003).

Usage

Use SpectralBiclustering when your data matrix has an inherent checkerboard structure, such as gene expression data where groups of genes respond similarly across groups of conditions. It is suitable when you want to simultaneously cluster both rows and columns rather than just one dimension.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/cluster/_bicluster.py

Signature

class SpectralBiclustering(BaseSpectral):
    def __init__(
        self,
        n_clusters=3,
        *,
        method="bistochastic",
        n_components=6,
        n_best=3,
        svd_method="randomized",
        n_svd_vecs=None,
        mini_batch=False,
        init="k-means++",
        n_init=10,
        random_state=None,
    ):

Import

from sklearn.cluster import SpectralBiclustering

I/O Contract

Inputs

Name	Type	Required	Description
n_clusters	int or tuple	No	Number of row and column clusters in the checkerboard structure. Default is 3.
method	str	No	Normalization method: "bistochastic", "scale", or "log". Default is "bistochastic".
n_components	int	No	Number of singular vectors to check. Default is 6.
n_best	int	No	Number of best singular vectors to use for clustering. Default is 3.
svd_method	str	No	SVD method: "randomized" or "arpack". Default is "randomized".
n_svd_vecs	int or None	No	Number of vectors for SVD computation. Default is None.
mini_batch	bool	No	Whether to use MiniBatchKMeans instead of KMeans. Default is False.
init	str	No	KMeans initialization method. Default is "k-means++".
n_init	int	No	Number of KMeans initializations. Default is 10.
random_state	int or RandomState	No	Random state for reproducibility. Default is None.

Outputs

Name	Type	Description
rows_	ndarray of shape (n_row_clusters, n_rows)	Boolean array indicating row membership in each bicluster.
columns_	ndarray of shape (n_column_clusters, n_columns)	Boolean array indicating column membership in each bicluster.
row_labels_	ndarray of shape (n_rows,)	Row cluster labels.
column_labels_	ndarray of shape (n_columns,)	Column cluster labels.

Usage Examples

Basic Usage

from sklearn.cluster import SpectralBiclustering
from sklearn.datasets import make_checkerboard
import numpy as np

data, rows, columns = make_checkerboard(
    shape=(300, 300), n_clusters=(4, 3), noise=10, random_state=0
)

model = SpectralBiclustering(n_clusters=3, random_state=0)
model.fit(data)
print(model.row_labels_)
print(model.column_labels_)

Related Pages

• Principle:Scikit_learn_Scikit_learn_Clustering