ai4he/randomized-spectral-co-clustering

Randomized Spectral Co-Clustering

Research code accompanying work on scalable spectral co-clustering using randomized matrix approximations. The repository contains MATLAB implementations of baseline spectral co-clustering and randomized variants based on random projection, random sampling, and related low-rank approximations.

Repository structure

• src/ — core co-clustering algorithms.
• src/support/ — supporting spectral-clustering and affinity-matrix routines.
• examples/ — runnable demonstrations and benchmark scripts.
• tests/ — lightweight smoke tests.
• utils/ — evaluation and data-loading utilities.
• data/ — small example datasets included for testing.
• docs/original_artifacts_manifest.csv — inventory of larger research artifacts omitted from GitHub.

Requirements

• MATLAB R2021a or newer is recommended.
• Statistics and Machine Learning Toolbox.

The code uses functions including kmeans, svds, eigs, spectralcluster, and confusionchart.

Quick start

Clone the repository, open MATLAB in the repository root, and run:

run('setup_project.m');
run('tests/run_smoke_tests.m');

To run a small comparison example:

run('examples/checkerboard_comparison.m');

Additional examples are available in the examples/ directory.

Included methods

The main implementations include:

• spectral_co_clustering.m — baseline spectral co-clustering.
• spectral_bi_clustering.m — spectral biclustering baseline.
• RP_SCC.m — random-projection spectral co-clustering.
• RPRE_SCC.m — randomized projection/reconstruction variant.
• RS_SCC.m — random-sampling spectral co-clustering.
• ID.m — interpolative-decomposition helper.

Data and reproducibility

Small datasets required by the smoke tests and selected examples are included under data/. Larger .mat datasets and MATLAB .fig files from the original experimental workspace are omitted to keep the repository lightweight. Their original inventory is provided in docs/original_artifacts_manifest.csv.

To reproduce experiments that require an omitted dataset, place the corresponding file in data/ and run the relevant script in examples/.

Because the randomized methods use stochastic projections, sampling, or clustering initialization, set the MATLAB random seed before a run when exact repeatability is needed:

rng(1, 'twister');

Citation

A machine-readable citation is provided in CITATION.cff. GitHub will display a Cite this repository button once the repository is public.

For now, the software may be cited as:

Mazdarani, F. (2026). Randomized Spectral Co-Clustering [Computer software]. GitHub.

When citing the research results, please cite the associated paper as well.

License

This project is released under the MIT License. See LICENSE.

Contact

For questions about the code or experiments, open a GitHub issue.

Dataset card

Paper	Randomized SVD Approximations for Spectral Co-Clustering of Word-Document Matrices
Venue	arXiv preprint arXiv:2609.19243, 2026
Authors	Fateme Mazdarani, Alberto Campos Hernandez, Carlos Toxtli
License	MIT License (mit)
Dataset DOI	10.5281/zenodo.24179048
Version	1.0.0
Code repository	https://github.com/mazdarani/randomized-spectral-co-clustering
Maintainer	Human-AI Empowerment Lab, Clemson University

Dataset structure

Path	Format	Size	Records
REPRODUCIBILITY.md	md	1.9 KB
setup_project.m	m	222 B
data/Checkerboard4x3_1000.mat	mat	7.6 MB
data/Checkerboard_100.mat	mat	80.1 KB
data/News4_wcsimp.mat	mat	338.1 KB
data/README.md	md	289 B
docs/original_artifacts_manifest.csv	csv	5.5 KB	86
examples/cancerdoc_demo.m	m	1.8 KB
examples/checkerboard_benchmark.m	m	2.7 KB
examples/checkerboard_comparison.m	m	4.1 KB
examples/news20_spectral_cluster_demo.m	m	639 B
examples/synthetic_coclustering_demo.m	m	1.5 KB
examples/synthetic_largescale_coclustering_demo.m	m	1.4 KB
src/ID.m	m	194 B
src/RPRE_SCC.m	m	1.1 KB
src/RP_SCC.m	m	996 B
src/RS_SCC.m	m	696 B
src/spectral_bi_clustering.m	m	501 B
src/spectral_co_clustering.m	m	780 B
src/support/cosineSimilarity.m	m	237 B
src/support/full_euclidean_affinity_matrix.m	m	358 B
src/support/knn_affinity_matrix.m	m	424 B
src/support/l1_affinity_matrix.m	m	242 B
src/support/laplacian.m	m	603 B
src/support/spectral_cluster.m	m	218 B
tests/run_smoke_tests.m	m	1.2 KB
utils/AccMeasure.m	m	1.4 KB
utils/require_data_file.m	m	389 B

28 files, 8.1 MB in total (metadata files excluded). Records are rows for CSV/TSV/JSONL files and list length for JSON arrays.

Loading the data

from datasets import load_dataset

ds = load_dataset("ai4he/randomized-spectral-co-clustering", "original_artifacts_manifest")

Available subsets: original_artifacts_manifest. All other files can be downloaded with huggingface_hub.snapshot_download or from the Zenodo record.

Known limitations

• Only small sample matrices are included; larger working files (e.g. CancerDoc.mat, 169 MB) are inventoried in docs/original_artifacts_manifest.csv but not distributed.

Personal and sensitive information

None.

License

The data and documentation are released under the MIT License.

Citation

If you use this dataset, please cite the paper:

@misc{Mazdarani2026Randomized,
 title = {Randomized SVD Approximations for Spectral Co-Clustering of Word-Document Matrices},
 author = {Mazdarani, Fateme and Toxtli, Carlos},
 year = {2026},
 eprint = {2609.19243},
 archiveprefix = {arXiv},
 primaryclass = {cs.LG},
 doi = {10.48550/arXiv.2609.19243},
 url = {https://arxiv.org/abs/2609.19243}
}

and the dataset:

@dataset{Mazdarani2026RandomizedSpectralCoClusData,
  author = {Mazdarani, Fateme and Campos Hernandez, Alberto and Toxtli, Carlos},
  title = {{Randomized Spectral Co-Clustering}: Dataset for ``Randomized SVD Approximations for Spectral Co-Clustering of Word-Document Matrices''},
  year = {2026},
  version = {1.0.0},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.24179048},
  url = {https://doi.org/10.5281/zenodo.24179048}
}