Generates tasks for training weights for the i'th label. More...

#include <training.h>

Inheritance diagram for dismec::TrainingTaskGenerator:

Public Member Functions
	TrainingTaskGenerator (std::shared_ptr< TrainingSpec > spec, label_id_t begin_label=label_id_t{0}, label_id_t end_label=label_id_t{-1})

	~TrainingTaskGenerator () override

void	run_tasks (long begin, long end, thread_id_t thread_id) override

void	prepare (long num_threads, long chunk_size) override
	Called to notify the `TaskGenerator` about the number of threads. More...

void	init_thread (thread_id_t thread_id) override
	Called once a thread has spun up, but before it runs its first task. More...

void	finalize () override
	Called after all threads have finished their tasks. More...

long	num_tasks () const override

const std::shared_ptr< model::Model > &	get_model () const

const std::vector< solvers::MinimizationResult > &	get_results () const

Public Member Functions inherited from dismec::parallel::TaskGenerator
virtual	~TaskGenerator ()=default

Private Member Functions
void	run_task (long task_id, thread_id_t thread_id)

solvers::MinimizationResult	train_label (label_id_t label_id, thread_id_t thread_id)
	Runs the training of a single label. More...

Private Attributes
std::shared_ptr< TrainingSpec >	m_TaskSpec

label_id_t	m_LabelRangeBegin

label_id_t	m_LabelRangeEnd

std::shared_ptr< model::Model >	m_Model

std::vector< solvers::MinimizationResult >	m_Results

std::vector< DenseRealVector >	m_ThreadLocalWorkingVector

std::vector< std::unique_ptr< solvers::Minimizer > >	m_ThreadLocalMinimizer

std::vector< std::shared_ptr< objective::Objective > >	m_ThreadLocalObjective

std::vector< std::unique_ptr< init::WeightsInitializer > >	m_ThreadLocalWeightInit

std::vector< std::unique_ptr< postproc::PostProcessor > >	m_ThreadLocalPostProc

std::vector< std::unique_ptr< ResultStatsGatherer > >	m_ResultGatherers

Additional Inherited Members
Public Types inherited from dismec::parallel::TaskGenerator
using	thread_id_t = dismec::parallel::thread_id_t

Detailed Description

Generates tasks for training weights for the i'th label.

This task generator produces tasks which train the weights for the different labels. The tasks are generated as specified by a TrainingSpec: Each thread generates a minimizer, an objective, and an initializer. The objective and optimizer are updated for each new label using the corresponding functions of the TrainingSpec object.

Definition at line 26 of file training.h.

Constructor & Destructor Documentation

◆ TrainingTaskGenerator()

TrainingTaskGenerator::TrainingTaskGenerator	(	std::shared_ptr< TrainingSpec >	spec,
		label_id_t	begin_label = `label_id_t{0}`,
		label_id_t	end_label = `label_id_t{-1}`
	)

explicit

Definition at line 20 of file training.cpp.

References m_LabelRangeBegin, m_LabelRangeEnd, m_Model, m_Results, and m_TaskSpec.

◆ ~TrainingTaskGenerator()

TrainingTaskGenerator::~TrainingTaskGenerator ( )

overridedefault

Member Function Documentation

◆ finalize()

void TrainingTaskGenerator::finalize ( )

overridevirtual

Called after all threads have finished their tasks.

This function is called from the main thread after all worker threads have finished their work. It can be used to perform single threaded reductions or clean up per-thread buffers.

Reimplemented from dismec::parallel::TaskGenerator.

Definition at line 108 of file training.cpp.

References m_TaskSpec, m_ThreadLocalMinimizer, m_ThreadLocalObjective, m_ThreadLocalPostProc, m_ThreadLocalWeightInit, and m_ThreadLocalWorkingVector.

◆ get_model()

const std::shared_ptr<model::Model>& dismec::TrainingTaskGenerator::get_model ( ) const

inline

Definition at line 38 of file training.h.

References m_Model.

◆ get_results()

const std::vector<solvers::MinimizationResult>& dismec::TrainingTaskGenerator::get_results ( ) const

inline

Definition at line 39 of file training.h.

References m_Results.

◆ init_thread()

void TrainingTaskGenerator::init_thread ( thread_id_t thread_id )

overridevirtual

Called once a thread has spun up, but before it runs its first task.

This function is called from inside the thread that also will run the tasks.

Reimplemented from dismec::parallel::TaskGenerator.

Definition at line 93 of file training.cpp.

References m_ResultGatherers, m_TaskSpec, m_ThreadLocalMinimizer, m_ThreadLocalObjective, m_ThreadLocalPostProc, m_ThreadLocalWeightInit, m_ThreadLocalWorkingVector, and dismec::opaque_int_type< Tag, T >::to_index().

◆ num_tasks()

long TrainingTaskGenerator::num_tasks ( ) const

overridevirtual

Implements dismec::parallel::TaskGenerator.

Definition at line 118 of file training.cpp.

References m_Results.

◆ prepare()

void TrainingTaskGenerator::prepare	(	long	num_threads,
		long	chunk_size
	)

overridevirtual

Called to notify the TaskGenerator about the number of threads.

This function is called from the main thread, before distributed work is started. It gives the TaskGenerator a chance to allocate working memory for each thread, so these allocations don't need to be done and repeated in run_task().

Note: For memory that is used inside the computations done by each thread, the init_thread() should be used. This will be called from inside the thread that will do the actual computations, so that when using this on a NUMA system, first-touch policy has a chance to place the allocation in the correct RAM. In that case, this function should only allocate an array of pointers that will be filled in by init_thread().

Parameters

num_threads	Number of threads that will be used.
chunk_size	A hint for the size of chunks used when running this task. Note that if the total number of tasks is not a multiple of the `chunk_size`, there may be some calls to `run_tasks()` with less than `chunk_size` tasks.

Reimplemented from dismec::parallel::TaskGenerator.

Definition at line 84 of file training.cpp.

References m_ResultGatherers, m_ThreadLocalMinimizer, m_ThreadLocalObjective, m_ThreadLocalPostProc, m_ThreadLocalWeightInit, and m_ThreadLocalWorkingVector.

◆ run_task()

void TrainingTaskGenerator::run_task	(	long	task_id,
		thread_id_t	thread_id
	)

private

Definition at line 43 of file training.cpp.

References m_LabelRangeBegin, m_Results, m_TaskSpec, dismec::opaque_int_type< Tag, T >::to_index(), and train_label().

Referenced by run_tasks().

◆ run_tasks()

void TrainingTaskGenerator::run_tasks	(	long	begin,
		long	end,
		thread_id_t	thread_id
	)

overridevirtual

Implements dismec::parallel::TaskGenerator.

Definition at line 37 of file training.cpp.

References run_task().

◆ train_label()

solvers::MinimizationResult TrainingTaskGenerator::train_label	(	label_id_t	label_id,
		thread_id_t	thread_id
	)

private

Runs the training of a single label.

Parameters

label_id	The id of the label which will be trained.
thread_id	The id of the thread on which training is running.

Returns: The return value of the minimizer.

Definition at line 50 of file training.cpp.

References m_Model, m_ResultGatherers, m_TaskSpec, m_ThreadLocalMinimizer, m_ThreadLocalObjective, m_ThreadLocalPostProc, m_ThreadLocalWeightInit, m_ThreadLocalWorkingVector, dismec::solvers::SUCCESS, and dismec::opaque_int_type< Tag, T >::to_index().

Referenced by run_task().

Member Data Documentation

◆ m_LabelRangeBegin

label_id_t dismec::TrainingTaskGenerator::m_LabelRangeBegin

private

Definition at line 56 of file training.h.

Referenced by run_task(), and TrainingTaskGenerator().

◆ m_LabelRangeEnd

label_id_t dismec::TrainingTaskGenerator::m_LabelRangeEnd

private

Definition at line 57 of file training.h.

Referenced by TrainingTaskGenerator().

◆ m_Model

std::shared_ptr<model::Model> dismec::TrainingTaskGenerator::m_Model

private

Definition at line 60 of file training.h.

Referenced by get_model(), train_label(), and TrainingTaskGenerator().

◆ m_ResultGatherers

std::vector<std::unique_ptr<ResultStatsGatherer> > dismec::TrainingTaskGenerator::m_ResultGatherers

private

Definition at line 69 of file training.h.

Referenced by init_thread(), prepare(), and train_label().

◆ m_Results

std::vector<solvers::MinimizationResult> dismec::TrainingTaskGenerator::m_Results

private

Definition at line 61 of file training.h.

Referenced by get_results(), num_tasks(), run_task(), and TrainingTaskGenerator().

◆ m_TaskSpec

std::shared_ptr<TrainingSpec> dismec::TrainingTaskGenerator::m_TaskSpec

private

Definition at line 53 of file training.h.

Referenced by finalize(), init_thread(), run_task(), train_label(), and TrainingTaskGenerator().

◆ m_ThreadLocalMinimizer

std::vector<std::unique_ptr<solvers::Minimizer> > dismec::TrainingTaskGenerator::m_ThreadLocalMinimizer

private

Definition at line 65 of file training.h.

Referenced by finalize(), init_thread(), prepare(), and train_label().

◆ m_ThreadLocalObjective

std::vector<std::shared_ptr<objective::Objective> > dismec::TrainingTaskGenerator::m_ThreadLocalObjective

private

Definition at line 66 of file training.h.

Referenced by finalize(), init_thread(), prepare(), and train_label().

◆ m_ThreadLocalPostProc

std::vector<std::unique_ptr<postproc::PostProcessor> > dismec::TrainingTaskGenerator::m_ThreadLocalPostProc

private

Definition at line 68 of file training.h.

Referenced by finalize(), init_thread(), prepare(), and train_label().

◆ m_ThreadLocalWeightInit

std::vector<std::unique_ptr<init::WeightsInitializer> > dismec::TrainingTaskGenerator::m_ThreadLocalWeightInit

private

Definition at line 67 of file training.h.

Referenced by finalize(), init_thread(), prepare(), and train_label().

◆ m_ThreadLocalWorkingVector

std::vector<DenseRealVector> dismec::TrainingTaskGenerator::m_ThreadLocalWorkingVector

private

Definition at line 64 of file training.h.

Referenced by finalize(), init_thread(), prepare(), and train_label().

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

src/training/training.h
src/training/training.cpp

Public Member Functions

Private Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ TrainingTaskGenerator()

◆ ~TrainingTaskGenerator()

Member Function Documentation

◆ finalize()

◆ get_model()

◆ get_results()

◆ init_thread()

◆ num_tasks()

◆ prepare()

◆ run_task()

◆ run_tasks()

◆ train_label()

Member Data Documentation

◆ m_LabelRangeBegin

◆ m_LabelRangeEnd

◆ m_Model

◆ m_ResultGatherers

◆ m_Results

◆ m_TaskSpec

◆ m_ThreadLocalMinimizer

◆ m_ThreadLocalObjective

◆ m_ThreadLocalPostProc

◆ m_ThreadLocalWeightInit

◆ m_ThreadLocalWorkingVector