This a simple K-Means algorithm for clustering points in the Poincare Disk Model, a model for hyperbolic space. This package was develop to exploit the results of Nickel & Kiela's Poincare Embeddings (as described in the paper Poincaré Embeddings for Learning Hierarchical Representations).
This code has not been optimized. All contributions (for example for optimizing it for large sample sizes) are welcome.
The API follows closely that of scikit-learn K-Means. A model is obtained by importing and instantiating PoincareKMeans
>>> model = PoincareKMeans()
The options are as follows.
- n_clusters (default 8): number of clusters to be determined
- n_init (default 20): number of time the k-means algorithm will be run with different centroid seeds.
- max_iter (default 300): maximum number of iterations of the k-means algorithm for a single run.
- tol (default 1e-8): tolerance criteria to declare convergence for each run.
- verbose (default True): verbosity mode. If True, will display the best inertia obtained for each run.
The model is trained on the dataset using fit
>>> model.fit(X)
Additional methods are provided:
- fit_predict: compute centroids and predict cluster index for each sample.
- fit_transform: compute clustering and transform X to cluster-distance space.
- predict: predict cluster index for the given sample.
- transform: computer cluster-distance for the given sample.
An example, using some coordinates obtained by Nickel & Kiela's embedding algorithm, is provided. The output should be analog to the following Figure.
