Base class for implementationa of an objective that combines dense features and sparse features. More...

#include <dense_and_sparse.h>

Inheritance diagram for dismec::objective::DenseAndSparseLinearBase:

Public Member Functions
	DenseAndSparseLinearBase (std::shared_ptr< const GenericFeatureMatrix > dense_features, std::shared_ptr< const GenericFeatureMatrix > sparse_features)

long	num_instances () const noexcept

long	num_variables () const noexcept override

BinaryLabelVector &	get_label_ref ()

void	update_costs (real_t positive, real_t negative)

void	update_features (const DenseFeatures &dense, const SparseFeatures &sparse)

Public Member Functions inherited from dismec::objective::Objective
	Objective ()

virtual	~Objective () noexcept=default

real_t	value (const HashVector &location)
	Evaluate the objective at the given `location`. More...

void	diag_preconditioner (const HashVector &location, Eigen::Ref< DenseRealVector > target)
	Get precondition to be used in CG optimization. More...

void	project_to_line (const HashVector &location, const DenseRealVector &direction)
	creates a function g such that `g(a) = objective(location + a * direction)` Use `lookup_on_line()` to evaluate `g`. More...

void	gradient_at_zero (Eigen::Ref< DenseRealVector > target)
	Gets the gradient for location zero. More...

void	gradient (const HashVector &location, Eigen::Ref< DenseRealVector > target)
	Evaluate the gradient at `location`. More...

void	hessian_times_direction (const HashVector &location, const DenseRealVector &direction, Eigen::Ref< DenseRealVector > target)
	Calculates the product of the Hessian matrix at `location` with `direction`. More...

void	gradient_and_pre_conditioner (const HashVector &location, Eigen::Ref< DenseRealVector > gradient, Eigen::Ref< DenseRealVector > pre)
	Combines the calculation of gradient and pre-conditioner, which may be more efficient in some cases. More...

Public Member Functions inherited from dismec::stats::Tracked
	Tracked ()
	Default constructor, creates the internal `stats::StatisticsCollection`. More...

void	register_stat (const std::string &name, std::unique_ptr< Statistics > stat)
	Registers a tracker for the statistics `name`. More...

std::shared_ptr< StatisticsCollection >	get_stats () const
	Gets an ownership-sharing reference to the `StatisticsCollection`. More...

Protected Member Functions
long	get_num_variables () const noexcept
	actual implementation of `num_variables()`. We need this non-virtual function to be called during the constructor More...

const DenseRealVector &	x_times_w (const HashVector &w)
	Calculates the vector of feature matrix times weights `w` More...

template<class Derived >
void	update_xtw_cache (const HashVector &new_weight, const Eigen::MatrixBase< Derived > &new_result)
	Updates the cached value for x_times_w. More...

void	project_linear_to_line (const HashVector &location, const DenseRealVector &direction)
	Prepares the cache variables for line projection. More...

auto	line_interpolation (real_t t) const

void	declare_vector_on_last_line (const HashVector &location, real_t t) override
	State that the given vector corresponds to a certain position on the line of the last line search. More...

const DenseFeatures &	dense_features () const

const SparseFeatures &	sparse_features () const

const DenseRealVector &	costs () const

const BinaryLabelVector &	labels () const

Protected Member Functions inherited from dismec::stats::Tracked
	~Tracked ()
	Non-virtual destructor. Declared protected, so we don't accidentally try to do a polymorphic delete. More...

template<class T >
void	record (stat_id_t stat, T &&value)
	Record statistics. This function just forwards all its arguments to the internal `StatisticsCollection`. More...

void	declare_stat (stat_id_t index, StatisticMetaData meta)
	Declares a new statistics. This function just forwards all its arguments to the internal `StatisticsCollection`. More...

void	declare_tag (tag_id_t index, std::string name)
	Declares a new tag. This function just forwards all its arguments to the internal `StatisticsCollection`. More...

template<class... Args>
void	set_tag (tag_id_t tag, long value)
	Set value of tag. This function just forwards all its arguments to the internal `StatisticsCollection`. More...

template<class... Args>
auto	make_timer (stat_id_t id, Args... args)
	Creates a new `ScopeTimer` using `stats::record_scope_time`. More...

Private Member Functions
real_t	value_unchecked (const HashVector &location) override

real_t	lookup_on_line (real_t position) override
	Looks up the value of the objective on the line defined by the last call to `project_to_line()`. More...

void	gradient_unchecked (const HashVector &location, Eigen::Ref< DenseRealVector > target) override

void	gradient_at_zero_unchecked (Eigen::Ref< DenseRealVector > target) override

void	hessian_times_direction_unchecked (const HashVector &location, const DenseRealVector &direction, Eigen::Ref< DenseRealVector > target) override

void	diag_preconditioner_unchecked (const HashVector &location, Eigen::Ref< DenseRealVector > target) override

void	gradient_and_pre_conditioner_unchecked (const HashVector &location, Eigen::Ref< DenseRealVector > gradient, Eigen::Ref< DenseRealVector > pre) override

void	project_to_line_unchecked (const HashVector &location, const DenseRealVector &direction) override

real_t	value_from_xTw (const DenseRealVector &xTw)

virtual void	calculate_loss (const DenseRealVector &scores, const BinaryLabelVector &labels, DenseRealVector &out) const =0

virtual void	calculate_derivative (const DenseRealVector &scores, const BinaryLabelVector &labels, DenseRealVector &out) const =0

virtual void	calculate_2nd_derivative (const DenseRealVector &scores, const BinaryLabelVector &labels, DenseRealVector &out) const =0

const DenseRealVector &	cached_derivative (const HashVector &location)

const DenseRealVector &	cached_2nd_derivative (const HashVector &location)

virtual real_t	regularization_value (const DenseRealVector &weights) const =0

virtual void	regularization_gradient (const DenseRealVector &weights, Eigen::Ref< DenseRealVector > gradient) const =0

virtual void	regularization_gradient_at_zero (Eigen::Ref< DenseRealVector > gradient) const =0

virtual void	regularization_preconditioner (const DenseRealVector &weights, Eigen::Ref< DenseRealVector > pre_cond) const =0

virtual void	regularization_hessian (const DenseRealVector &weights, const DenseRealVector &direction, Eigen::Ref< DenseRealVector > pre_cond) const =0

void	invalidate_labels ()

Private Attributes
std::shared_ptr< const GenericFeatureMatrix >	m_DenseFeatures
	Pointer to the dense part of the feature matrix. More...

std::shared_ptr< const GenericFeatureMatrix >	m_SparseFeatures
	pointer to the sparse part of the feature matrix More...

VectorHash	m_Last_W {}
	cache for the last argument to `x_times_w()`. More...

DenseRealVector	m_X_times_w
	cache for the last result of `x_times_w()` corresponding to `m_Last_W`. More...

DenseRealVector	m_LsCache_xTd
	cache for line search implementation: feature times direction More...

DenseRealVector	m_LsCache_xTw
	cache for line search implementation: feature times weights More...

DenseRealVector	m_Costs
	Label-Dependent costs. More...

BinaryLabelVector	m_Y
	Label vector – use a vector of ints here. We encode label present == 1, absent == -1. More...

CacheHelper	m_DerivativeBuffer

CacheHelper	m_SecondDerivativeBuffer

DenseRealVector	m_LineStart

DenseRealVector	m_LineDirection

DenseRealVector	m_LineCache

DenseRealVector	m_GenericInBuffer

DenseRealVector	m_GenericOutBuffer

Detailed Description

Base class for implementationa of an objective that combines dense features and sparse features.

This is the shared code for all combined sparse/dense feature linear objectives.

Note: Unfortunately, I don't think that this can be implemented as a subclass of the Linear base, even though it models a linear classifier.

Definition at line 20 of file dense_and_sparse.h.

Constructor & Destructor Documentation

◆ DenseAndSparseLinearBase()

DenseAndSparseLinearBase::DenseAndSparseLinearBase	(	std::shared_ptr< const GenericFeatureMatrix >	dense_features,
		std::shared_ptr< const GenericFeatureMatrix >	sparse_features
	)

Definition at line 20 of file dense_and_sparse.cpp.

References ALWAYS_ASSERT_EQUAL, dismec::stats::Tracked::declare_stat(), m_Costs, m_DenseFeatures, m_SparseFeatures, and anonymous_namespace{dense_and_sparse.cpp}::STAT_PERF_MATMUL.

Member Function Documentation

◆ cached_2nd_derivative()

const DenseRealVector & DenseAndSparseLinearBase::cached_2nd_derivative ( const HashVector & location )

private

Definition at line 209 of file dense_and_sparse.cpp.

References calculate_2nd_derivative(), costs(), labels(), m_SecondDerivativeBuffer, dismec::CacheHelper::update(), and x_times_w().

Referenced by diag_preconditioner_unchecked(), gradient_and_pre_conditioner_unchecked(), and hessian_times_direction_unchecked().

◆ cached_derivative()

const DenseRealVector & DenseAndSparseLinearBase::cached_derivative ( const HashVector & location )

private

Definition at line 201 of file dense_and_sparse.cpp.

References calculate_derivative(), costs(), labels(), m_DerivativeBuffer, dismec::CacheHelper::update(), and x_times_w().

Referenced by gradient_and_pre_conditioner_unchecked(), and gradient_unchecked().

◆ calculate_2nd_derivative()

virtual void dismec::objective::DenseAndSparseLinearBase::calculate_2nd_derivative	(	const DenseRealVector &	scores,
		const BinaryLabelVector &	labels,
		DenseRealVector &	out
	)		const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by cached_2nd_derivative().

◆ calculate_derivative()

virtual void dismec::objective::DenseAndSparseLinearBase::calculate_derivative	(	const DenseRealVector &	scores,
		const BinaryLabelVector &	labels,
		DenseRealVector &	out
	)		const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by cached_derivative(), and gradient_at_zero_unchecked().

◆ calculate_loss()

virtual void dismec::objective::DenseAndSparseLinearBase::calculate_loss	(	const DenseRealVector &	scores,
		const BinaryLabelVector &	labels,
		DenseRealVector &	out
	)		const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by value_from_xTw().

◆ costs()

const DenseRealVector & DenseAndSparseLinearBase::costs ( ) const

protected

Definition at line 98 of file dense_and_sparse.cpp.

References m_Costs.

Referenced by cached_2nd_derivative(), cached_derivative(), gradient_at_zero_unchecked(), and value_from_xTw().

◆ declare_vector_on_last_line()

void dismec::objective::DenseAndSparseLinearBase::declare_vector_on_last_line	(	const HashVector &	location,
		real_t	t
	)

inlineoverrideprotectedvirtual

State that the given vector corresponds to a certain position on the line of the last line search.

This function is a pure optimization hint. It is used in the following scenario: If several computations need the product of weight vector w and feature matrix X, then we can compute this product only once and use a cached value for all later invocations. This can be done by comparing the vector hashes. However, as soon as a vector is modified, these hashes are invalidated. To do an efficient line search over ‘w’ = w + t d, we also cache the value ofX d, so thatX w' = X w + t X d. This function then declares that the vector given inlocationcorresponds tow + t d, wherewandd are the arguments passed to the last call ofproject_to_line()`.

Todo:: improve this interface, together with project_to_line, to be less error prone!

Reimplemented from dismec::objective::Objective.

Definition at line 75 of file dense_and_sparse.h.

References m_LsCache_xTd, m_LsCache_xTw, and update_xtw_cache().

◆ dense_features()

const DenseFeatures & DenseAndSparseLinearBase::dense_features ( ) const

protected

Definition at line 51 of file dense_and_sparse.cpp.

References m_DenseFeatures.

Referenced by diag_preconditioner_unchecked(), gradient_and_pre_conditioner_unchecked(), gradient_at_zero_unchecked(), gradient_unchecked(), hessian_times_direction_unchecked(), project_linear_to_line(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_gradient(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_gradient_at_zero(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_hessian(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_preconditioner(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_value(), and x_times_w().

◆ diag_preconditioner_unchecked()

void DenseAndSparseLinearBase::diag_preconditioner_unchecked	(	const HashVector &	location,
		Eigen::Ref< DenseRealVector >	target
	)

overrideprivatevirtual

The function that does the actual computation. This is called in diag_preconditioner() after the arguments have been validated. The default implementation returns ones.

Reimplemented from dismec::objective::Objective.

Definition at line 189 of file dense_and_sparse.cpp.

References cached_2nd_derivative(), dense_features(), DENSE_PART, dismec::HashVector::get(), regularization_preconditioner(), sparse_features(), and SPARSE_PART.

◆ get_label_ref()

BinaryLabelVector & DenseAndSparseLinearBase::get_label_ref ( )

Definition at line 81 of file dense_and_sparse.cpp.

References invalidate_labels(), and m_Y.

◆ get_num_variables()

long DenseAndSparseLinearBase::get_num_variables ( ) const

protectednoexcept

actual implementation of num_variables(). We need this non-virtual function to be called during the constructor

Definition at line 47 of file dense_and_sparse.cpp.

References m_DenseFeatures, and m_SparseFeatures.

Referenced by num_variables().

◆ gradient_and_pre_conditioner_unchecked()

void DenseAndSparseLinearBase::gradient_and_pre_conditioner_unchecked	(	const HashVector &	location,
		Eigen::Ref< DenseRealVector >	gradient,
		Eigen::Ref< DenseRealVector >	pre
	)

overrideprivatevirtual

The function that does the actual computation. This is called in gradient_and_pre_conditioner() after the arguments have been validated. The default implementation sucessively calls gradient() and diag_preconditioner().

Reimplemented from dismec::objective::Objective.

Definition at line 141 of file dense_and_sparse.cpp.

References cached_2nd_derivative(), cached_derivative(), dense_features(), DENSE_PART, dismec::HashVector::get(), dismec::objective::Objective::gradient(), regularization_gradient(), regularization_preconditioner(), sparse_features(), and SPARSE_PART.

◆ gradient_at_zero_unchecked()

void DenseAndSparseLinearBase::gradient_at_zero_unchecked ( Eigen::Ref< DenseRealVector > target )

overrideprivatevirtual

The function that does the actual computation. This is called in gradient_at_zero() after the argument has been validated. The default implementation is rather inefficient and creates a new temporary zero vector.

Reimplemented from dismec::objective::Objective.

Definition at line 174 of file dense_and_sparse.cpp.

References calculate_derivative(), costs(), dense_features(), DENSE_PART, labels(), m_GenericInBuffer, m_GenericOutBuffer, regularization_gradient_at_zero(), sparse_features(), and SPARSE_PART.

◆ gradient_unchecked()

void DenseAndSparseLinearBase::gradient_unchecked	(	const HashVector &	location,
		Eigen::Ref< DenseRealVector >	target
	)

overrideprivatevirtual

The function that does the actual gradient computation. This is called in gradient() after the arguments have been validated.

Implements dismec::objective::Objective.

Definition at line 162 of file dense_and_sparse.cpp.

References cached_derivative(), dense_features(), DENSE_PART, dismec::HashVector::get(), regularization_gradient(), sparse_features(), and SPARSE_PART.

◆ hessian_times_direction_unchecked()

void DenseAndSparseLinearBase::hessian_times_direction_unchecked	(	const HashVector &	location,
		const DenseRealVector &	direction,
		Eigen::Ref< DenseRealVector >	target
	)

overrideprivatevirtual

The function that does the actual computation. This is called in hessian_times_direction() after the arguments have been validated.

Implements dismec::objective::Objective.

Definition at line 126 of file dense_and_sparse.cpp.

References cached_2nd_derivative(), dense_features(), DENSE_PART, dismec::HashVector::get(), regularization_hessian(), sparse_features(), and SPARSE_PART.

◆ invalidate_labels()

void DenseAndSparseLinearBase::invalidate_labels ( )

private

This function will be called whenever m_Y changes so that derived classes can invalidate their caches.

Definition at line 217 of file dense_and_sparse.cpp.

References dismec::CacheHelper::invalidate(), m_DerivativeBuffer, and m_SecondDerivativeBuffer.

Referenced by get_label_ref().

◆ labels()

const BinaryLabelVector & DenseAndSparseLinearBase::labels ( ) const

protected

Definition at line 102 of file dense_and_sparse.cpp.

References m_Y.

Referenced by cached_2nd_derivative(), cached_derivative(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::calculate_2nd_derivative(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::calculate_derivative(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::calculate_loss(), gradient_at_zero_unchecked(), update_costs(), and value_from_xTw().

◆ line_interpolation()

auto dismec::objective::DenseAndSparseLinearBase::line_interpolation ( real_t t ) const

inlineprotected

Definition at line 71 of file dense_and_sparse.h.

References m_LsCache_xTd, and m_LsCache_xTw.

Referenced by lookup_on_line().

◆ lookup_on_line()

real_t DenseAndSparseLinearBase::lookup_on_line ( real_t position )

overrideprivatevirtual

Looks up the value of the objective on the line defined by the last call to project_to_line().

Parameters

position The location where the objective is calculated.

Returns: The value of objective(location + position * direction), where location and direction are the vectors passed to the last call of project_to_line().

Attention: This function may use results cached project_to_line(), so it has to be called after a call to that function. A new call to project_to_line() will change the line which is evaluated.

Implements dismec::objective::Objective.

Definition at line 111 of file dense_and_sparse.cpp.

References line_interpolation(), m_GenericInBuffer, m_LineCache, m_LineDirection, m_LineStart, regularization_value(), and value_from_xTw().

◆ num_instances()

long DenseAndSparseLinearBase::num_instances ( ) const

noexcept

Definition at line 39 of file dense_and_sparse.cpp.

References m_DenseFeatures.

◆ num_variables()

long DenseAndSparseLinearBase::num_variables ( ) const

overridevirtualnoexcept

Gets the number of variables this objective expects. May return -1 if the objective is agnostic to the number of variables, e.g. for regularizers.

Implements dismec::objective::Objective.

Definition at line 43 of file dense_and_sparse.cpp.

References get_num_variables().

◆ project_linear_to_line()

void DenseAndSparseLinearBase::project_linear_to_line	(	const HashVector &	location,
		const DenseRealVector &	direction
	)

protected

Prepares the cache variables for line projection.

This function precomputes \( x^T d\) and \(x^T w\), so that for a line search parameter t we can get \( x^T (w + td) \) by simple linear combination, i.e. we can skip the matrix multiplication. This can then be used in the implementation of the line search lookup by calling the line_interpolation() function.

Parameters

location	The origin point on the line.
direction	The direction of the line.

Definition at line 73 of file dense_and_sparse.cpp.

References dense_features(), DENSE_PART, dismec::HashVector::get(), m_LineDirection, m_LineStart, m_LsCache_xTd, m_LsCache_xTw, sparse_features(), SPARSE_PART, and x_times_w().

Referenced by project_to_line_unchecked().

◆ project_to_line_unchecked()

void DenseAndSparseLinearBase::project_to_line_unchecked	(	const HashVector &	location,
		const DenseRealVector &	direction
	)

overrideprivatevirtual

The function that does the actual computation. This is called in project_to_line() after the arguments have been validated.

Implements dismec::objective::Objective.

Definition at line 222 of file dense_and_sparse.cpp.

References project_linear_to_line().

◆ regularization_gradient()

virtual void dismec::objective::DenseAndSparseLinearBase::regularization_gradient	(	const DenseRealVector &	weights,
		Eigen::Ref< DenseRealVector >	gradient
	)		const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by gradient_and_pre_conditioner_unchecked(), and gradient_unchecked().

◆ regularization_gradient_at_zero()

virtual void dismec::objective::DenseAndSparseLinearBase::regularization_gradient_at_zero ( Eigen::Ref< DenseRealVector > gradient ) const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by gradient_at_zero_unchecked().

◆ regularization_hessian()

virtual void dismec::objective::DenseAndSparseLinearBase::regularization_hessian	(	const DenseRealVector &	weights,
		const DenseRealVector &	direction,
		Eigen::Ref< DenseRealVector >	pre_cond
	)		const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by hessian_times_direction_unchecked().

◆ regularization_preconditioner()

virtual void dismec::objective::DenseAndSparseLinearBase::regularization_preconditioner	(	const DenseRealVector &	weights,
		Eigen::Ref< DenseRealVector >	pre_cond
	)		const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by diag_preconditioner_unchecked(), and gradient_and_pre_conditioner_unchecked().

◆ regularization_value()

virtual real_t dismec::objective::DenseAndSparseLinearBase::regularization_value ( const DenseRealVector & weights ) const

privatepure virtual

Implemented in dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >.

Referenced by lookup_on_line(), and value_unchecked().

◆ sparse_features()

const SparseFeatures & DenseAndSparseLinearBase::sparse_features ( ) const

protected

Definition at line 55 of file dense_and_sparse.cpp.

References m_SparseFeatures.

Referenced by diag_preconditioner_unchecked(), gradient_and_pre_conditioner_unchecked(), gradient_at_zero_unchecked(), gradient_unchecked(), hessian_times_direction_unchecked(), project_linear_to_line(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_gradient(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_gradient_at_zero(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_hessian(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_preconditioner(), dismec::objective::DenseAndSparseMargin< MarginFunction, SparseRegFunction, DenseRegFunction >::regularization_value(), and x_times_w().

◆ update_costs()

void DenseAndSparseLinearBase::update_costs	(	real_t	positive,
		real_t	negative
	)

Definition at line 87 of file dense_and_sparse.cpp.

References labels(), m_Costs, and m_Y.

◆ update_features()

void DenseAndSparseLinearBase::update_features	(	const DenseFeatures &	dense,
		const SparseFeatures &	sparse
	)

Definition at line 226 of file dense_and_sparse.cpp.

References m_DenseFeatures, and m_SparseFeatures.

◆ update_xtw_cache()

template<class Derived >

void dismec::objective::DenseAndSparseLinearBase::update_xtw_cache	(	const HashVector &	new_weight,
		const Eigen::MatrixBase< Derived > &	new_result
	)

inlineprotected

Updates the cached value for x_times_w.

Parameters

new_weight	The new value of w.
new_result	The value of x^T w.

Definition at line 52 of file dense_and_sparse.h.

References dismec::HashVector::hash(), m_Last_W, and m_X_times_w.

Referenced by declare_vector_on_last_line().

◆ value_from_xTw()

real_t DenseAndSparseLinearBase::value_from_xTw ( const DenseRealVector & xTw )

private

Definition at line 118 of file dense_and_sparse.cpp.

References calculate_loss(), costs(), labels(), and m_GenericOutBuffer.

Referenced by lookup_on_line(), and value_unchecked().

◆ value_unchecked()

real_t DenseAndSparseLinearBase::value_unchecked ( const HashVector & location )

overrideprivatevirtual

The function that does the actual value computation. This is called in value() after the argument has been validated.

Implements dismec::objective::Objective.

Definition at line 106 of file dense_and_sparse.cpp.

References dismec::HashVector::get(), regularization_value(), value_from_xTw(), and x_times_w().

◆ x_times_w()

const DenseRealVector & DenseAndSparseLinearBase::x_times_w ( const HashVector & w )

protected

Calculates the vector of feature matrix times weights w

Parameters

w	The weight vector with which to multiply.

Consecutive calls to this function with the same argument w will return a reference to a cached result. However, calling with another value in between will invalidate the cache.

Returns: A reference to the cached result vector.

Definition at line 62 of file dense_and_sparse.cpp.

References dense_features(), DENSE_PART, dismec::HashVector::get(), dismec::HashVector::hash(), m_Last_W, m_X_times_w, dismec::stats::Tracked::make_timer(), sparse_features(), SPARSE_PART, and anonymous_namespace{dense_and_sparse.cpp}::STAT_PERF_MATMUL.

Referenced by cached_2nd_derivative(), cached_derivative(), project_linear_to_line(), and value_unchecked().

Member Data Documentation

◆ m_Costs

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_Costs

private

Label-Dependent costs.

Definition at line 144 of file dense_and_sparse.h.

Referenced by costs(), DenseAndSparseLinearBase(), and update_costs().

◆ m_DenseFeatures

std::shared_ptr<const GenericFeatureMatrix> dismec::objective::DenseAndSparseLinearBase::m_DenseFeatures

private

Pointer to the dense part of the feature matrix.

Definition at line 129 of file dense_and_sparse.h.

Referenced by dense_features(), DenseAndSparseLinearBase(), get_num_variables(), num_instances(), and update_features().

◆ m_DerivativeBuffer

CacheHelper dismec::objective::DenseAndSparseLinearBase::m_DerivativeBuffer

private

Definition at line 152 of file dense_and_sparse.h.

Referenced by cached_derivative(), and invalidate_labels().

◆ m_GenericInBuffer

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_GenericInBuffer

private

Definition at line 159 of file dense_and_sparse.h.

Referenced by gradient_at_zero_unchecked(), and lookup_on_line().

◆ m_GenericOutBuffer

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_GenericOutBuffer

private

Definition at line 160 of file dense_and_sparse.h.

Referenced by gradient_at_zero_unchecked(), and value_from_xTw().

◆ m_Last_W

VectorHash dismec::objective::DenseAndSparseLinearBase::m_Last_W {}

private

cache for the last argument to x_times_w().

Definition at line 134 of file dense_and_sparse.h.

Referenced by update_xtw_cache(), and x_times_w().

◆ m_LineCache

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_LineCache

private

Definition at line 157 of file dense_and_sparse.h.

Referenced by lookup_on_line().

◆ m_LineDirection

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_LineDirection

private

Definition at line 156 of file dense_and_sparse.h.

Referenced by lookup_on_line(), and project_linear_to_line().

◆ m_LineStart

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_LineStart

private

Definition at line 155 of file dense_and_sparse.h.

Referenced by lookup_on_line(), and project_linear_to_line().

◆ m_LsCache_xTd

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_LsCache_xTd

private

cache for line search implementation: feature times direction

Definition at line 139 of file dense_and_sparse.h.

Referenced by declare_vector_on_last_line(), line_interpolation(), and project_linear_to_line().

◆ m_LsCache_xTw

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_LsCache_xTw

private

cache for line search implementation: feature times weights

Definition at line 141 of file dense_and_sparse.h.

Referenced by declare_vector_on_last_line(), line_interpolation(), and project_linear_to_line().

◆ m_SecondDerivativeBuffer

CacheHelper dismec::objective::DenseAndSparseLinearBase::m_SecondDerivativeBuffer

private

Definition at line 153 of file dense_and_sparse.h.

Referenced by cached_2nd_derivative(), and invalidate_labels().

◆ m_SparseFeatures

std::shared_ptr<const GenericFeatureMatrix> dismec::objective::DenseAndSparseLinearBase::m_SparseFeatures

private

pointer to the sparse part of the feature matrix

Definition at line 131 of file dense_and_sparse.h.

Referenced by DenseAndSparseLinearBase(), get_num_variables(), sparse_features(), and update_features().

◆ m_X_times_w

DenseRealVector dismec::objective::DenseAndSparseLinearBase::m_X_times_w

private

cache for the last result of x_times_w() corresponding to m_Last_W.

Definition at line 136 of file dense_and_sparse.h.

Referenced by update_xtw_cache(), and x_times_w().

◆ m_Y

BinaryLabelVector dismec::objective::DenseAndSparseLinearBase::m_Y

private

Label vector – use a vector of ints here. We encode label present == 1, absent == -1.

Definition at line 147 of file dense_and_sparse.h.

Referenced by get_label_ref(), labels(), and update_costs().

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

src/objective/dense_and_sparse.h
src/objective/dense_and_sparse.cpp

Public Member Functions

Protected Member Functions

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ DenseAndSparseLinearBase()

Member Function Documentation

◆ cached_2nd_derivative()

◆ cached_derivative()

◆ calculate_2nd_derivative()

◆ calculate_derivative()

◆ calculate_loss()

◆ costs()

◆ declare_vector_on_last_line()

◆ dense_features()

◆ diag_preconditioner_unchecked()

◆ get_label_ref()

◆ get_num_variables()

◆ gradient_and_pre_conditioner_unchecked()

◆ gradient_at_zero_unchecked()

◆ gradient_unchecked()

◆ hessian_times_direction_unchecked()

◆ invalidate_labels()

◆ labels()

◆ line_interpolation()

◆ lookup_on_line()

◆ num_instances()

◆ num_variables()

◆ project_linear_to_line()

◆ project_to_line_unchecked()

◆ regularization_gradient()

◆ regularization_gradient_at_zero()

◆ regularization_hessian()

◆ regularization_preconditioner()

◆ regularization_value()

◆ sparse_features()

◆ update_costs()

◆ update_features()

◆ update_xtw_cache()

◆ value_from_xTw()

◆ value_unchecked()

◆ x_times_w()

Member Data Documentation

◆ m_Costs

◆ m_DenseFeatures

◆ m_DerivativeBuffer

◆ m_GenericInBuffer

◆ m_GenericOutBuffer

◆ m_Last_W

◆ m_LineCache

◆ m_LineDirection

◆ m_LineStart

◆ m_LsCache_xTd

◆ m_LsCache_xTw

◆ m_SecondDerivativeBuffer

◆ m_SparseFeatures

◆ m_X_times_w

◆ m_Y