umappp Namespace Reference

Methods for UMAP. More...

Classes
struct	Options
	Options for initialize(). More...
class	Status
	Status of the UMAP optimization iterations. More...

Typedefs
template<typename Index_ , typename Float_ >
using	NeighborList = knncolle::NeighborList< Index_, Float_ >
	Lists of neighbors for each observation.

Enumerations
enum	InitializeMethod : char { SPECTRAL , SPECTRAL_ONLY , RANDOM , NONE }

Functions
template<typename Index_ , typename Float_ >
Status< Index_, Float_ >	initialize (NeighborList< Index_, Float_ > x, int num_dim, Float_ *embedding, Options options)
template<typename Dim_ , typename Index_ , typename Float_ >
Status< Index_, Float_ >	initialize (const knncolle::Prebuilt< Dim_, Index_, Float_ > &prebuilt, int num_dim, Float_ *embedding, Options options)
template<typename Dim_ , typename Index_ , typename Float_ >
Status< Index_, Float_ >	initialize (Dim_ data_dim, Index_ num_obs, const Float_ *data, const knncolle::Builder< knncolle::SimpleMatrix< Dim_, Index_, Float_ >, Float_ > &builder, int num_dim, Float_ *embedding, Options options)

Detailed Description

Methods for UMAP.

Typedef Documentation

NeighborList

template<typename Index_ , typename Float_ >

using umappp::NeighborList = typedef knncolle::NeighborList<Index_, Float_>

Lists of neighbors for each observation.

Template Parameters

Index_	Integer type of the neighbor indices.
Float_	Floating-point type for the distances.

This is a convenient alias for the knncolle::NeighborList class. Each inner vector corresponds to an observation and contains the list of nearest neighbors for that observation, sorted by increasing distance. Neighbors for each observation should be unique - there should be no more than one occurrence of each index in each inner vector. Also, the inner vector for observation i should not contain any Neighbor with index i.

Enumeration Type Documentation

InitializeMethod

enum umappp::InitializeMethod : char

How should the initial coordinates of the embedding be obtained?

• SPECTRAL: attempts initialization based on spectral decomposition of the graph Laplacian. If that fails, we fall back to random draws from a normal distribution.
• SPECTRAL_ONLY: attempts spectral initialization as before, but if that fails, we use the existing values in the supplied embedding array.
• RANDOM: fills the embedding with random draws from a normal distribution.
• NONE: uses the existing values in the supplied embedding array.

Function Documentation

initialize() [1/3]

template<typename Dim_ , typename Index_ , typename Float_ >

Status< Index_, Float_ > umappp::initialize	(	const knncolle::Prebuilt< Dim_, Index_, Float_ > &	prebuilt,
		int	num_dim,
		Float_ *	embedding,
		Options	options
	)

Template Parameters

Dim_	Integer type for the dimensions of the input dataset.
Index_	Integer type of the neighbor indices.
Float_	Floating-point type for the distances.

Parameters

	prebuilt	A knncolle::Prebuilt instance constructed from the input dataset.
	num_dim	Number of dimensions of the UMAP embedding.
[in,out]	embedding	Pointer to an array in which to store the embedding, where rows are dimensions (num_dim) and columns are observations (x.size()). This is only used as input if Options::init == InitializeMethod::NONE, otherwise it is only used as output. The lifetime of the array should be no shorter than the final call to Status::run().
	options	Further options.

Returns

A Status object containing the initial state of the UMAP algorithm. Further calls to Status::run() will update the embeddings in embedding.

initialize() [2/3]

template<typename Dim_ , typename Index_ , typename Float_ >

Status< Index_, Float_ > umappp::initialize	(	Dim_	data_dim,
		Index_	num_obs,
		const Float_ *	data,
		const knncolle::Builder< knncolle::SimpleMatrix< Dim_, Index_, Float_ >, Float_ > &	builder,
		int	num_dim,
		Float_ *	embedding,
		Options	options
	)

Template Parameters

Dim_	Integer type for the dimensions of the input dataset.
Index_	Integer type of the neighbor indices.
Float_	Floating-point type for the distances.

Parameters

	data_dim	Number of dimensions of the input dataset.
	num_obs	Number of observations in the input dataset.
[in]	data	Pointer to an array containing the input high-dimensional data as a column-major matrix. Each row corresponds to a dimension (data_dim) and each column corresponds to an observation (num_obs).
	builder	Algorithm to use for the neighbor search.
	num_dim	Number of dimensions of the embedding.
[in,out]	embedding	Pointer to an array in which to store the embedding, where rows are dimensions (num_dim) and columns are observations (x.size()). This is only used as input if Options::init == InitializeMethod::NONE, otherwise it is only used as output. The lifetime of the array should be no shorter than the final call to Status::run().
	options	Further options.

Returns

A Status object containing the initial state of the UMAP algorithm. Further calls to Status::run() will update the embeddings in embedding.

initialize() [3/3]

template<typename Index_ , typename Float_ >

Status< Index_, Float_ > umappp::initialize	(	NeighborList< Index_, Float_ >	x,
		int	num_dim,
		Float_ *	embedding,
		Options	options
	)

Template Parameters

Index_	Integer type of the neighbor indices.
Float_	Floating-point type for the distances.

Parameters

	x	Indices and distances to the nearest neighbors for each observation. Note the expectations in the NeighborList documentation.
	num_dim	Number of dimensions of the embedding.
[in,out]	embedding	Pointer to an array in which to store the embedding, where rows are dimensions (num_dim) and columns are observations (x.size()). This is only used as input if Options::init == InitializeMethod::NONE, otherwise it is only used as output. The lifetime of the array should be no shorter than the final call to Status::run().
	options	Further options. Note that Options::num_neighbors is ignored here.

Returns

A Status object containing the initial state of the UMAP algorithm. Further calls to Status::run() will update the embeddings in embedding.