C++ implementation of Dbscan algo

Fits a DBSCAN model on an input feature matrix and outputs the labels and core_sample_indices. More...

Functions
void	ML::Dbscan::fit (const raft::handle_t &handle, float *input, int n_rows, int n_cols, float eps, int min_pts, cuvs::distance::DistanceType metric, int *labels, int *core_sample_indices=nullptr, float *sample_weight=nullptr, size_t max_bytes_per_batch=0, EpsNnMethod eps_nn_method=BRUTE_FORCE, int verbosity=CUML_LEVEL_INFO, bool opg=false)
void	ML::Dbscan::fit (const raft::handle_t &handle, double *input, int n_rows, int n_cols, double eps, int min_pts, cuvs::distance::DistanceType metric, int *labels, int *core_sample_indices=nullptr, double *sample_weight=nullptr, size_t max_bytes_per_batch=0, EpsNnMethod eps_nn_method=BRUTE_FORCE, int verbosity=CUML_LEVEL_INFO, bool opg=false)
void	ML::Dbscan::fit (const raft::handle_t &handle, float *input, int64_t n_rows, int64_t n_cols, float eps, int min_pts, cuvs::distance::DistanceType metric, int64_t *labels, int64_t *core_sample_indices=nullptr, float *sample_weight=nullptr, size_t max_bytes_per_batch=0, EpsNnMethod eps_nn_method=BRUTE_FORCE, int verbosity=CUML_LEVEL_INFO, bool opg=false)
void	ML::Dbscan::fit (const raft::handle_t &handle, double *input, int64_t n_rows, int64_t n_cols, double eps, int min_pts, cuvs::distance::DistanceType metric, int64_t *labels, int64_t *core_sample_indices=nullptr, double *sample_weight=nullptr, size_t max_bytes_per_batch=0, EpsNnMethod eps_nn_method=BRUTE_FORCE, int verbosity=CUML_LEVEL_INFO, bool opg=false)

Detailed Description

Fits a DBSCAN model on an input feature matrix and outputs the labels and core_sample_indices.

Parameters

[in]	handle	cuml handle to use across the algorithm
[in]	input	row-major input feature matrix or distance matrix
[in]	n_rows	number of samples in the input feature matrix
[in]	n_cols	number of features in the input feature matrix
[in]	eps	epsilon value to use for epsilon-neighborhood determination
[in]	min_pts	minimum number of points to determine a cluster
[in]	metric	metric type (or precomputed)
[out]	labels	(size n_rows) output labels array
[out]	core_sample_indices	(size n_rows) output array containing the indices of each core point. If the number of core points is less than n_rows, the right will be padded with -1. Setting this to NULL will prevent calculating the core sample indices
[in]	sample_weight	(size n_rows) input array containing the weight of each sample to be taken instead of a plain sum to fulfill the min_pts criteria for core points. NULL will default to weights of 1 for all samples
[in]	max_bytes_per_batch	the maximum number of megabytes to be used for each batch of the pairwise distance calculation. This enables the trade off between memory usage and algorithm execution time.
[in]	eps_nn_method	method for computing epsilon neighborhood
[in]	verbosity	verbosity level for logging messages during execution
[in]	opg	whether we are running in a multi-node multi-GPU context

Function Documentation

fit() [1/4]

void ML::Dbscan::fit	(	const raft::handle_t &	handle,
		double *	input,
		int	n_rows,
		int	n_cols,
		double	eps,
		int	min_pts,
		cuvs::distance::DistanceType	metric,
		int *	labels,
		int *	core_sample_indices = nullptr,
		double *	sample_weight = nullptr,
		size_t	max_bytes_per_batch = 0,
		EpsNnMethod	eps_nn_method = BRUTE_FORCE,
		int	verbosity = CUML_LEVEL_INFO,
		bool	opg = false
	)

fit() [2/4]

void ML::Dbscan::fit	(	const raft::handle_t &	handle,
		double *	input,
		int64_t	n_rows,
		int64_t	n_cols,
		double	eps,
		int	min_pts,
		cuvs::distance::DistanceType	metric,
		int64_t *	labels,
		int64_t *	core_sample_indices = nullptr,
		double *	sample_weight = nullptr,
		size_t	max_bytes_per_batch = 0,
		EpsNnMethod	eps_nn_method = BRUTE_FORCE,
		int	verbosity = CUML_LEVEL_INFO,
		bool	opg = false
	)

fit() [3/4]

void ML::Dbscan::fit	(	const raft::handle_t &	handle,
		float *	input,
		int	n_rows,
		int	n_cols,
		float	eps,
		int	min_pts,
		cuvs::distance::DistanceType	metric,
		int *	labels,
		int *	core_sample_indices = nullptr,
		float *	sample_weight = nullptr,
		size_t	max_bytes_per_batch = 0,
		EpsNnMethod	eps_nn_method = BRUTE_FORCE,
		int	verbosity = CUML_LEVEL_INFO,
		bool	opg = false
	)

fit() [4/4]

void ML::Dbscan::fit	(	const raft::handle_t &	handle,
		float *	input,
		int64_t	n_rows,
		int64_t	n_cols,
		float	eps,
		int	min_pts,
		cuvs::distance::DistanceType	metric,
		int64_t *	labels,
		int64_t *	core_sample_indices = nullptr,
		float *	sample_weight = nullptr,
		size_t	max_bytes_per_batch = 0,
		EpsNnMethod	eps_nn_method = BRUTE_FORCE,
		int	verbosity = CUML_LEVEL_INFO,
		bool	opg = false
	)