optimizes the order of indices in a ProductQuantizer More...

Inheritance diagram for faiss::PolysemousTraining:

Public Types
enum	Optimization_type_t { OT_None, OT_ReproduceDistances_affine, OT_Ranking_weighted_diff }

Public Member Functions
void	optimize_pq_for_hamming (ProductQuantizer &pq, size_t n, const float *x) const

void	optimize_ranking (ProductQuantizer &pq, size_t n, const float *x) const
	called by optimize_pq_for_hamming

void	optimize_reproduce_distances (ProductQuantizer &pq) const
	called by optimize_pq_for_hamming

Public Attributes
Optimization_type_t	optimization_type

int	ntrain_permutation

double	dis_weight_factor

std::string	log_pattern

Public Attributes inherited from faiss::SimulatedAnnealingParameters
double	init_temperature

double	temperature_decay

int	n_iter

int	n_redo

int	seed

int	verbose

bool	only_bit_flips

bool	init_random

Detailed Description

optimizes the order of indices in a ProductQuantizer

Definition at line 127 of file PolysemousTraining.h.

Member Enumeration Documentation

Enumerator
OT_ReproduceDistances_affine	default
OT_Ranking_weighted_diff	same as _2, but use rank of y+ - rank of y-

Definition at line 129 of file PolysemousTraining.h.

void faiss::PolysemousTraining::optimize_pq_for_hamming	(	ProductQuantizer &	pq,
		size_t	n,
		const float *	x
	)		const

reorder the centroids so that the Hamming distace becomes a good approximation of the SDC distance (called by train)

Definition at line 945 of file PolysemousTraining.cpp.

The documentation for this struct was generated from the following files: