knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ > Class Template Reference

Wrapper around a builder with L2 normalization. More...

#include <L2Normalized.hpp>

Inheritance diagram for knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >:

Collaboration diagram for knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >:

Public Types
typedef L2NormalizedMatrix< Index_, Data_, Normalized_, Matrix_ >	NormalizedMatrix
typedef std::conditional< std::is_base_of< Matrix_, NormalizedMatrix >::value, Matrix_, NormalizedMatrix >::type	BuilderMatrix

Public Member Functions
	L2NormalizedBuilder (std::shared_ptr< const Builder< Index_, Normalized_, Distance_, BuilderMatrix > > builder)
Prebuilt< Index_, Data_, Distance_ > *	build_raw (const Matrix_ &data) const
auto	build_known_raw (const Matrix_ &data) const
auto	build_known_unique (const Matrix_ &data) const
auto	build_known_shared (const Matrix_ &data) const
Public Member Functions inherited from knncolle::Builder< Index_, Data_, Distance_, Matrix_ >
std::shared_ptr< Prebuilt< Index_, Data_, Distance_ > >	build_shared (const Matrix_ &data) const
std::unique_ptr< Prebuilt< Index_, Data_, Distance_ > >	build_unique (const Matrix_ &data) const
auto	build_known_raw (const Matrix_ &data) const
auto	build_known_unique (const Matrix_ &data) const
auto	build_known_shared (const Matrix_ &data) const

Detailed Description

template<typename Index_, typename Data_, typename Distance_, typename Normalized_, class Matrix_ = Matrix<Index_, Data_>>
class knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >

Wrapper around a builder with L2 normalization.

This class applies L2 normalization to each observation vector in its input matrix, and also constructs Searcher instances that L2-normalize each query vector. The premise is that Euclidean distances on L2-normalized vectors are monotonic transformations of the cosine distance. Thus, given an arbitrary algorithm that finds nearest neighbors according to Euclidean distance, users can wrap the former's Builder with this L2NormalizedBuilder to obtain neighbors according to the cosine distance.

Template Parameters

Index_	Integer type for the indices.
Data_	Numeric type for the input and query data.
Distance_	Floating point type for the distances.
Normalized_	Floating-point type for the L2-normalized data.
Matrix_	Class of the input data matrix. This should satisfy the Matrix interface.

Member Typedef Documentation

BuilderMatrix

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

std::conditional<std::is_base_of<Matrix_,NormalizedMatrix>::value,Matrix_,NormalizedMatrix>::type knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::BuilderMatrix

The expected matrix type in the Builder instance in the constructor.

If Matrix_ is a base class of NormalizedMatrix, the Builder can use the same Matrix_ in its template parametrization. This is because Matrix_ will be compatible with both the L2NormalizedMatrix and the type of the original matrix, allowing us to transparently substitute an instance of the latter with that of the former. In general, this scenario requires the default Matrix_ type and Data_ == Normalized_.

Otherwise, the Builder should explicitly have a NormalizedMatrix in its template parametrization. This is because it will be accepting the normalized matrix in build_raw() instead of Matrix_.

NormalizedMatrix

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

L2NormalizedMatrix<Index_, Data_, Normalized_, Matrix_> knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::NormalizedMatrix

Alias for the type of the normalized matrix.

Constructor & Destructor Documentation

L2NormalizedBuilder()

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::L2NormalizedBuilder ( std::shared_ptr< const Builder< Index_, Normalized_, Distance_, BuilderMatrix > > builder )

inline

Parameters

builder

Pointer to a Builder for an arbitrary neighbor search algorithm.

Member Function Documentation

build_known_raw()

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

auto knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::build_known_raw ( const Matrix_ & data ) const

inline

Override to assist devirtualization.

build_known_shared()

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

auto knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::build_known_shared ( const Matrix_ & data ) const

inline

Override to assist devirtualization.

build_known_unique()

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

auto knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::build_known_unique ( const Matrix_ & data ) const

inline

Override to assist devirtualization.

build_raw()

template<typename Index_ , typename Data_ , typename Distance_ , typename Normalized_ , class Matrix_ = Matrix<Index_, Data_>>

Prebuilt< Index_, Data_, Distance_ > * knncolle::L2NormalizedBuilder< Index_, Data_, Distance_, Normalized_, Matrix_ >::build_raw ( const Matrix_ & data ) const

inlinevirtual

Parameters

data	Object satisfying the Matrix interface, containing observations in columns and dimensions in rows.

Returns

Pointer to a pre-built search index.

Implements knncolle::Builder< Index_, Data_, Distance_, Matrix_ >.

---

The documentation for this class was generated from the following file:

• knncolle/L2Normalized.hpp