ML::Spectral Namespace Reference

Functions
void	fit_embedding (const raft::handle_t &handle, int *rows, int *cols, float *vals, int nnz, int n, int n_components, float *out, unsigned long long seed=1234567)

Function Documentation

fit_embedding()

void ML::Spectral::fit_embedding	(	const raft::handle_t &	handle,
		int *	rows,
		int *	cols,
		float *	vals,
		int	nnz,
		int	n,
		int	n_components,
		float *	out,
		unsigned long long	seed = 1234567
	)

Given a COO formatted (symmetric) knn graph, this function computes the spectral embeddings (lowest n_components eigenvectors), using Lanczos min cut algorithm.

Parameters

handle	cuml handle
rows	source vertices of knn graph (size nnz)
cols	destination vertices of knn graph (size nnz)
vals	edge weights connecting vertices of knn graph (size nnz)
nnz	size of rows/cols/vals
n	number of samples in X
n_components	the number of components to project the X into
out	output array for embedding (size n*n_comonents)
seed	random seed to use in both the lanczos solver and k-means