ML::HDBSCAN Namespace Reference

Namespaces
	Common
	detail
	HELPER

Classes
struct	FixConnectivitiesRedOp

Functions
template<typename value_idx = int64_t, typename value_t = float>
void	build_linkage (const raft::handle_t &handle, const value_t *X, size_t m, size_t n, cuvs::distance::DistanceType metric, Common::HDBSCANParams &params, value_t *core_dists, Common::robust_single_linkage_output< value_idx, value_t > &out)
template<typename value_idx = int64_t, typename value_t = float>
void	_fit_hdbscan (const raft::handle_t &handle, const value_t *X, size_t m, size_t n, cuvs::distance::DistanceType metric, Common::HDBSCANParams &params, value_idx *labels, value_t *core_dists, Common::hdbscan_output< value_idx, value_t > &out)

Function Documentation

_fit_hdbscan()

template<typename value_idx = int64_t, typename value_t = float>

void ML::HDBSCAN::_fit_hdbscan	(	const raft::handle_t &	handle,
		const value_t *	X,
		size_t	m,
		size_t	n,
		cuvs::distance::DistanceType	metric,
		Common::HDBSCANParams &	params,
		value_idx *	labels,
		value_t *	core_dists,
		Common::hdbscan_output< value_idx, value_t > &	out
	)

Condense branches of tree according to min cluster size

Extract labels from stability

Normalize labels so they are drawn from a monotonically increasing set starting at 0 even in the presence of noise (-1)

build_linkage()

template<typename value_idx = int64_t, typename value_t = float>

void ML::HDBSCAN::build_linkage	(	const raft::handle_t &	handle,
		const value_t *	X,
		size_t	m,
		size_t	n,
		cuvs::distance::DistanceType	metric,
		Common::HDBSCANParams &	params,
		value_t *	core_dists,
		Common::robust_single_linkage_output< value_idx, value_t > &	out
	)

Constructs a linkage by computing mutual reachability, mst, and dendrogram. This is shared by HDBSCAN and Robust Single Linkage since the two algorithms differ only in the cluster selection and extraction.

Template Parameters

value_idx
value_t

Parameters

[in]	handle	raft handle for resource reuse
[in]	X	data points (size m * n)
[in]	m	number of rows
[in]	n	number of columns
[in]	metric	distance metric to use
[in]	params	hyper parameters
[in]	core_dists	buffer for storing core distances (size m)
[out]	out	output container object

Mutual reachability graph

Construct MST sorted by weights

Perform hierarchical labeling