PyPI version License: MIT Python Versions Paper

G-CoMVKM is a Python implementation of the Globally Collaborative Multi-View k-Means clustering algorithm. This algorithm integrates a collaborative transfer learning framework with entropy-regularized feature-view reduction, enabling dynamic elimination of uninformative components. The method achieves clustering by balancing local view importance and global consensus.

• Multi-View Clustering: Process data from multiple views/sources simultaneously
• Feature Weight Learning: Automatically determine the importance of each feature
• View Weight Learning: Automatically determine the importance of each view
• Feature Selection: Entropy-regularized mechanism to discard irrelevant features
• Global Consensus: Balance local view objectives with global clustering agreement

You can install G-CoMVKM directly from PyPI:

pip install gcomvkm

• Python 3.7+
• NumPy
• SciPy
• Matplotlib
• scikit-learn
• seaborn

Here's a simple example of how to use G-CoMVKM:

from gcomvkm import GCoMVKM
from gcomvkm.utils import load_synthetic_data
from gcomvkm.evaluation import nmi, rand_index, adjusted_rand_index
# Load the synthetic dataset (2 views, 2 dimensions, 2 clusters)
X, true_labels = load_synthetic_data()
# Create and fit the model
model = GCoMVKM(
 n_clusters=2,
 gamma=5.0, # Feature selection regularization parameter
 theta=4.0, # View weight regularization parameter
 max_iter=100,
 tol=1e-4,
 verbose=True,
 random_state=42
)
# Fit the model to the data
model.fit(X)
# Get the clustering results
predicted_labels = model.labels_
feature_weights = model.feature_weights_
view_weights = model.view_weights_
# Evaluate clustering performance
nmi_score = nmi(true_labels, predicted_labels)
ri_score = rand_index(true_labels, predicted_labels)
ari_score = adjusted_rand_index(true_labels, predicted_labels)
print(f"NMI Score: {nmi_score:.4f}")
print(f"Rand Index: {ri_score:.4f}")
print(f"Adjusted Rand Index: {ari_score:.4f}")

G-CoMVKM extends the traditional k-means algorithm to work with multi-view data. The algorithm:

1. Initializes cluster centers randomly or using k-means++
2. Computes memberships for each data point to the clusters
3. Updates cluster centers based on these memberships
4. Updates feature weights using an entropy-regularized optimization
5. Discards irrelevant features based on a threshold
6. Updates view weights to balance view importance
7. Repeats steps 2-6 until convergence

The objective function minimizes the within-cluster variance while encouraging feature and view sparsity through entropy regularization.

G-CoMVKM also provides a command-line interface:

# Run with default parameters on the synthetic dataset
gcomvkm --dataset 2V2D2C
# Run with custom parameters
gcomvkm --dataset 2V2D2C --gamma 5.0 --theta 4.0 --n-clusters 2 --max-iter 100

1. Comprehensive Cross-Platform Development

   • ✅ Production-grade MATLAB Implementation (original repository)
   • ✅ Professional Python Package (PyPI: gcomvkm 0.1.0)
   • ✅ Industry-standard documentation and interactive tutorials
   • ✅ 100% reproducible experiments with provided code and data
   • ✅ Optimized performance with GPU acceleration

2. Quality Assurance

   • Rigorous testing across multiple datasets
   • Comprehensive error handling and input validation
   • Performance benchmarking against state-of-the-art methods
   • Clean, well-documented, and maintainable code

3. User Experience

   • Intuitive API design following scikit-learn conventions
   • Detailed documentation with examples and tutorials
   • Visualizations for better interpretation of results
   • Command-line interface for quick experimentation

If you use G-CoMVKM in your research, please cite:

@Article{electronics14112129,
AUTHOR = {Sinaga, Kristina P. and Yang, Miin-Shen},
TITLE = {A Globally Collaborative Multi-View k-Means Clustering},
JOURNAL = {Electronics},
VOLUME = {14},
YEAR = {2025},
NUMBER = {11},
ARTICLE-NUMBER = {2129},
URL = {https://www.mdpi.com/2079-9292/14/11/2129},
ISSN = {2079-9292},
ABSTRACT = {Multi-view (MV) data are increasingly collected from various fields, like IoT. The surge in MV data demands clustering algorithms capable of handling heterogeneous features and high dimensionality. Existing feature-weighted MV k-means (MVKM) algorithms often neglect effective dimensionality reduction such that their scalability and interpretability are limited. To address this, we propose a novel procedure for clustering MV data, namely a globally collaborative MVKM (G-CoMVKM) clustering algorithm. The proposed G-CoMVKM integrates a collaborative transfer learning framework with entropy-regularized feature-view reduction, enabling dynamic elimination of uninformative components. This method achieves clustering by balancing local view importance and global consensus, without relying on matrix reconstruction. We design a feature-view reduction by embedding transferred learning processes across view components by using penalty terms and entropy to simultaneously reduce these unimportant feature-view components. Experiments on synthetic and real-world datasets demonstrate that G-CoMVKM consistently outperforms these existing MVKM clustering algorithms in clustering accuracy, performance, and dimensionality reduction, affirming its robustness and efficiency.},
DOI = {10.3390/electronics14112129}
}

The original code has been tested on MATLAB R2020a. Performance on other versions may vary. This Python implementation has been tested on Python 3.7+ and is compatible with most modern Python environments.

As Arthur C. Clarke said, "The only way of discovering the limits of the possible is to venture a little way past them into the impossible."

We didn't just venture—we blazed a trail:

• Where they saw complexity, we found elegance
• Where they predicted failure, we achieved excellence
• Where they set limits, we broke boundaries
• Where they said "impossible," we said "watch us"

To aspiring researchers: Let our journey be a reminder that in science, "impossible" is often just a challenge waiting to be accepted. The boundaries of what's possible are meant to be pushed, tested, and ultimately redefined.

• Kristina P. Sinaga
• Email: kristinapestaria.sinaga@isti.cnr.it (The email address kristinasinaga41@gmail.com is no longer under my authority. Please do not use it to contact me).
• GitHub

1. A Globally Collaborative Multi-View k-Means Clustering - Electronics MDPI
2. Original MATLAB Implementation: G-CoMVKM

This project is licensed under the MIT License - see the LICENSE file for details.

Contributions are welcome! Please feel free to submit a Pull Request.