SOM Plus Clustering
An extended self-organizing map library with smarter initialization options and built-in cluster quality metrics — designed to make unsupervised topology learning more rigorous and reproducible.
Overview
What is it?
Standard SOM implementations leave too much to chance — random initialization can produce wildly different maps across runs, and most libraries provide no way to evaluate whether the resulting clusters are actually meaningful.
SOM Plus extends the base algorithm with PCA-guided initialization for deterministic, data-aware starting weights, alongside silhouette score and Davies-Bouldin index evaluation so you can objectively compare runs and configurations. It grew out of the GDP trajectory clustering project and the functional group analysis work, where reliable clustering was critical to interpretation.
Features
What it does
Extended SOM
Full self-organizing map implementation with configurable topology, neighbourhood functions, and learning rate decay.
PCA Initialization
Seeds neuron weights along principal components of the data — faster convergence and reproducible maps.
Random Initialization
Traditional random init still available with seeding for exact reproducibility across experiments.
Silhouette Score
Measures how well-separated clusters are — higher is better, and you get it automatically after training.
Davies-Bouldin Index
Ratio of within-cluster scatter to between-cluster separation — lower scores mean tighter, more distinct clusters.
Research-ready
Designed for scientific use — clean API, NumPy-native, and compatible with SciPy analysis pipelines.
Stack
Built with
See the code
Full source, examples, and API docs on GitHub.