k_nearest_neighbors

tangles.util.graph.similarity.k_nearest_neighbors(X: ndarray, metric: str = 'precomputed', k: int = 1, ties_all: bool = False) → csr_matrix

Creates a k-nearest neighbor graph (or something like a k-nearest neighbor graph) from distances.

Parameters

Xnp.ndarray

If metric is ‘precomputed’ X is a condensed distance matrix (see scipy.spatial.distance.squareform()). Otherwise X is the data and scipy.spatial.distance.pdist() is called taking X as its argument.

metricstr

Either ‘precomputed’ or a metric from scipy.spatial.distance.pdist(). If not ‘precomputed’, metric is forwarded to the scipy.spatial.distance.pdist() call.

kint

Number of neighbors.

ties_allbool

If True, every node gets a connection to all other nodes that have a distance smaller than the k-nearest neighbor, i.e. if there are multiple k-nearest neighbors, we connect them all, resulting in a graph that might have degrees greater than k (so it is not really a k-nn graph, but something similar where ties are broken in a special way). If False, only one of possibly multiple k-nearest neighbors is connected.

Returns

sparse.csr_matrix

Sparse adjacency matrix of a k-nn graph (note that this graph is directed).