topo.layouts.isomap

Functions

Isomap(X[, n_components, n_neighbors, metric, ...])

Isomap embedding of a dataset or precomputed graph.

Module Contents

topo.layouts.isomap.Isomap(X, n_components=2, n_neighbors=50, metric='cosine', landmarks=None, landmark_method='kmeans', eig_tol=0, n_jobs=1, **kwargs)

Isomap embedding of a dataset or precomputed graph.

Computes geodesic distances on the kNN graph, applies double-centering to the squared distance matrix (classical MDS kernel), and returns the top eigenvectors scaled by the square root of the corresponding eigenvalues — the standard Isomap procedure.

Parameters:

  • X (array-like or sparse) – Input data matrix of shape (n_samples, n_features), or a precomputed kNN graph (sparse) when metric='precomputed'.

  • n_components (int, default 2) – Number of dimensions to embed into.

  • n_neighbors (int, default 50) – Number of neighbors used to build the kNN graph (ignored when metric='precomputed').

  • metric (str, default 'cosine') – Distance metric for the kNN graph. Pass 'precomputed' if X is already a sparse adjacency matrix.

  • landmarks (int or array-like of int, optional) – If an int, the number of landmarks to select (using landmark_method). If an array, the explicit landmark indices. When set, only the landmark-to-landmark geodesic sub-matrix is used.

  • landmark_method ({'kmeans', 'random'}, default 'kmeans') – Method for landmark selection when landmarks is an integer.

  • eig_tol (float, default 0) – Stopping tolerance for the eigendecomposition (passed to ARPACK).

  • n_jobs (int, default 1) – Number of parallel jobs for kNN construction and shortest-path computation.

  • **kwargs – Additional keyword arguments forwarded to topo.base.ann.kNN().

Returns:

Y (ndarray of shape (n_samples, n_components)) – Isomap embedding coordinates.