template<typename Architecture_t>
class TMVA::DNN::TDLGradientDescent< Architecture_t >

Definition at line 65 of file DLMinimizers.h.

Public Types
using	DeepNet_t = TDeepNet< Architecture_t >

using	Matrix_t = typename Architecture_t::Matrix_t

using	Scalar_t = typename Architecture_t::Scalar_t

Public Member Functions
	TDLGradientDescent ()

	TDLGradientDescent (Scalar_t learningRate, size_t convergenceSteps, size_t testInterval)

size_t	GetConvergenceCount () const
	Getters.

size_t	GetConvergenceSteps () const

Scalar_t	GetTestError () const

size_t	GetTestInterval () const

Scalar_t	GetTrainingError () const

bool	HasConverged ()
	Increases the minimization step counter by the test error evaluation period and uses the current internal value of the test error to determine if the minimization has converged.

bool	HasConverged (Scalar_t testError)
	Increases the minimization step counter by the test error evaluation period and uses the provided test error value to determine if the minimization has converged.

void	Reset ()
	Reset minimizer object to default state.

void	SetBatchSize (Scalar_t rate)

void	SetConvergenceSteps (size_t steps)
	Setters.

void	SetLearningRate (Scalar_t rate)

void	SetTestInterval (size_t interval)

void	Step (DeepNet_t &deepNet, std::vector< Matrix_t > &input, const Matrix_t &output, const Matrix_t &weights)
	Perform a single optimization step on a given batch.

void	Step (DeepNet_t &master, std::vector< DeepNet_t > &nets, std::vector< TTensorBatch< Architecture_t > > &batches)
	Perform multiple optimization steps simultaneously.

Scalar_t	StepLoss (DeepNet_t &deepNet, std::vector< Matrix_t > &input, const Matrix_t &output, const Matrix_t &weights)
	Same as Step(...) but also evaluate the loss on the given training data.

void	StepMomentum (DeepNet_t &master, std::vector< DeepNet_t > &nets, std::vector< TTensorBatch< Architecture_t > > &batches, Scalar_t momentum)
	Same as the Step(...) method for multiple batches but uses momentum.

void	StepNesterov (DeepNet_t &master, std::vector< DeepNet_t > &nets, std::vector< TTensorBatch< Architecture_t > > &batches, Scalar_t momentum)
	Same as the Step(...) method for multiple batches but uses Nesterov momentum.

void	StepReducedWeights (DeepNet_t &deepNet, std::vector< Matrix_t > &input, const Matrix_t &output, const Matrix_t &weights)
	Does not evaluate the loss and therefore not trigger a possible synchronization with the device.

Scalar_t	StepReducedWeightsLoss (DeepNet_t &deepNet, std::vector< Matrix_t > &input, const Matrix_t &output, const Matrix_t &weights)
	Similar to StepReducedWeights(...) but also evaluates the loss.

Private Attributes
size_t	fBatchSize
	Batch size to use for the training.

size_t	fConvergenceCount
	Current number of training epochs without.

size_t	fConvergenceSteps
	Number of training epochs without considerable.

Scalar_t	fLearningRate
	Learning rate \(\alpha\).

Scalar_t	fMinimumError
	The minimum loss achieved on the training set during the current training session.

size_t	fStepCount
	Number of steps performed in the current training session.

Scalar_t	fTestError
	Holds the most recently computed test loss.

size_t	fTestInterval
	Interval for the computation of the test error.

Scalar_t	fTrainingError
	Holds the most recently computed training loss.

#include <TMVA/DNN/DLMinimizers.h>

Member Typedef Documentation

◆ DeepNet_t

template<typename Architecture_t >

using TMVA::DNN::TDLGradientDescent< Architecture_t >::DeepNet_t = TDeepNet<Architecture_t>

Definition at line 67 of file DLMinimizers.h.

◆ Matrix_t

template<typename Architecture_t >

using TMVA::DNN::TDLGradientDescent< Architecture_t >::Matrix_t = typename Architecture_t::Matrix_t

Definition at line 69 of file DLMinimizers.h.

◆ Scalar_t

template<typename Architecture_t >

using TMVA::DNN::TDLGradientDescent< Architecture_t >::Scalar_t = typename Architecture_t::Scalar_t

Definition at line 68 of file DLMinimizers.h.

Constructor & Destructor Documentation

◆ TDLGradientDescent() [1/2]

template<typename Architecture_t >

TMVA::DNN::TDLGradientDescent< Architecture_t >::TDLGradientDescent

Definition at line 163 of file DLMinimizers.h.

◆ TDLGradientDescent() [2/2]

template<typename Architecture_t >

TMVA::DNN::TDLGradientDescent< Architecture_t >::TDLGradientDescent	(	Scalar_t	learningRate,
		size_t	convergenceSteps,
		size_t	testInterval
	)

Definition at line 172 of file DLMinimizers.h.

Member Function Documentation

◆ GetConvergenceCount()

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::GetConvergenceCount ( ) const

inline

Getters.

Definition at line 146 of file DLMinimizers.h.

◆ GetConvergenceSteps()

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::GetConvergenceSteps ( ) const

inline

Definition at line 147 of file DLMinimizers.h.

◆ GetTestError()

template<typename Architecture_t >

Scalar_t TMVA::DNN::TDLGradientDescent< Architecture_t >::GetTestError ( ) const

inline

Definition at line 149 of file DLMinimizers.h.

◆ GetTestInterval()

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::GetTestInterval ( ) const

inline

Definition at line 150 of file DLMinimizers.h.

◆ GetTrainingError()

template<typename Architecture_t >

Scalar_t TMVA::DNN::TDLGradientDescent< Architecture_t >::GetTrainingError ( ) const

inline

Definition at line 148 of file DLMinimizers.h.

◆ HasConverged() [1/2]

template<typename Architecture_t >

bool TMVA::DNN::TDLGradientDescent< Architecture_t >::HasConverged

Increases the minimization step counter by the test error evaluation period and uses the current internal value of the test error to determine if the minimization has converged.

Definition at line 276 of file DLMinimizers.h.

◆ HasConverged() [2/2]

template<typename Architecture_t >

bool TMVA::DNN::TDLGradientDescent< Architecture_t >::HasConverged ( Scalar_t testError )

Increases the minimization step counter by the test error evaluation period and uses the provided test error value to determine if the minimization has converged.

Definition at line 290 of file DLMinimizers.h.

◆ Reset()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::Reset ( )

inline

Reset minimizer object to default state.

Definition at line 89 of file DLMinimizers.h.

◆ SetBatchSize()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::SetBatchSize ( Scalar_t rate )

inline

Definition at line 156 of file DLMinimizers.h.

◆ SetConvergenceSteps()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::SetConvergenceSteps ( size_t steps )

inline

Setters.

Definition at line 153 of file DLMinimizers.h.

◆ SetLearningRate()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::SetLearningRate ( Scalar_t rate )

inline

Definition at line 155 of file DLMinimizers.h.

◆ SetTestInterval()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::SetTestInterval ( size_t interval )

inline

Definition at line 154 of file DLMinimizers.h.

◆ Step() [1/2]

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::Step	(	DeepNet_t &	deepNet,
		std::vector< Matrix_t > &	input,
		const Matrix_t &	output,
		const Matrix_t &	weights
	)

Perform a single optimization step on a given batch.

Propagates the input matrix forward through the net, evaluates the loss and propagates the gradients backward through the net. The computed gradients are scaled by the learning rate \(\alpha\) and subtracted from the weights and bias values of each layer.

Definition at line 182 of file DLMinimizers.h.

◆ Step() [2/2]

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::Step	(	DeepNet_t &	master,
		std::vector< DeepNet_t > &	nets,
		std::vector< TTensorBatch< Architecture_t > > &	batches
	)

Perform multiple optimization steps simultaneously.

Performs the backprop algorithm on the input batches given in batches on the neural networks given in nets. The forward and backward propagation steps are executed in an interleaving manner in order to exploit potential batch-level parallelism for asynchronous device calls.

Definition at line 246 of file DLMinimizers.h.

◆ StepLoss()

template<typename Architecture_t >

auto TMVA::DNN::TDLGradientDescent< Architecture_t >::StepLoss	(	DeepNet_t &	deepNet,
		std::vector< Matrix_t > &	input,
		const Matrix_t &	output,
		const Matrix_t &	weights
	)

Same as Step(...) but also evaluate the loss on the given training data.

Note that this requires synchronization between host and device.

Definition at line 212 of file DLMinimizers.h.

◆ StepMomentum()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::StepMomentum	(	DeepNet_t &	master,
		std::vector< DeepNet_t > &	nets,
		std::vector< TTensorBatch< Architecture_t > > &	batches,
		Scalar_t	momentum
	)

Same as the Step(...) method for multiple batches but uses momentum.

Definition at line 256 of file DLMinimizers.h.

◆ StepNesterov()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::StepNesterov	(	DeepNet_t &	master,
		std::vector< DeepNet_t > &	nets,
		std::vector< TTensorBatch< Architecture_t > > &	batches,
		Scalar_t	momentum
	)

Same as the Step(...) method for multiple batches but uses Nesterov momentum.

Definition at line 266 of file DLMinimizers.h.

◆ StepReducedWeights()

template<typename Architecture_t >

void TMVA::DNN::TDLGradientDescent< Architecture_t >::StepReducedWeights	(	DeepNet_t &	deepNet,
		std::vector< Matrix_t > &	input,
		const Matrix_t &	output,
		const Matrix_t &	weights
	)

Does not evaluate the loss and therefore not trigger a possible synchronization with the device.

Trains the weights of each layer, but only the bias terms of the first layer for compatibility with the previous implementation.

Definition at line 193 of file DLMinimizers.h.

◆ StepReducedWeightsLoss()

template<typename Architecture_t >

auto TMVA::DNN::TDLGradientDescent< Architecture_t >::StepReducedWeightsLoss	(	DeepNet_t &	deepNet,
		std::vector< Matrix_t > &	input,
		const Matrix_t &	output,
		const Matrix_t &	weights
	)

Similar to StepReducedWeights(...) but also evaluates the loss.

May trigger synchronization with the device.

Definition at line 224 of file DLMinimizers.h.

Member Data Documentation

◆ fBatchSize

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fBatchSize

private

Batch size to use for the training.

Definition at line 72 of file DLMinimizers.h.

◆ fConvergenceCount

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fConvergenceCount

private

Current number of training epochs without.

considerable decrease in the test error.

Definition at line 76 of file DLMinimizers.h.

◆ fConvergenceSteps

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fConvergenceSteps

private

Number of training epochs without considerable.

decrease in the test error for convergence.

Definition at line 74 of file DLMinimizers.h.

◆ fLearningRate

template<typename Architecture_t >

Scalar_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fLearningRate

private

Learning rate \(\alpha\).

Definition at line 81 of file DLMinimizers.h.

◆ fMinimumError

template<typename Architecture_t >

Scalar_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fMinimumError

private

The minimum loss achieved on the training set during the current training session.

Definition at line 82 of file DLMinimizers.h.

◆ fStepCount

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fStepCount

private

Number of steps performed in the current training session.

Definition at line 73 of file DLMinimizers.h.

◆ fTestError

template<typename Architecture_t >

Scalar_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fTestError

private

Holds the most recently computed test loss.

Definition at line 80 of file DLMinimizers.h.

◆ fTestInterval

template<typename Architecture_t >

size_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fTestInterval

private

Interval for the computation of the test error.

Definition at line 78 of file DLMinimizers.h.

◆ fTrainingError

template<typename Architecture_t >

Scalar_t TMVA::DNN::TDLGradientDescent< Architecture_t >::fTrainingError

private

Holds the most recently computed training loss.

Definition at line 79 of file DLMinimizers.h.

tmva/tmva/inc/TMVA/DNN/DLMinimizers.h

Public Types

Public Member Functions

Private Attributes

Member Typedef Documentation

◆ DeepNet_t

◆ Matrix_t

◆ Scalar_t

Constructor & Destructor Documentation

◆ TDLGradientDescent() [1/2]

◆ TDLGradientDescent() [2/2]

Member Function Documentation

◆ GetConvergenceCount()

◆ GetConvergenceSteps()

◆ GetTestError()

◆ GetTestInterval()

◆ GetTrainingError()

◆ HasConverged() [1/2]

◆ HasConverged() [2/2]

◆ Reset()

◆ SetBatchSize()

◆ SetConvergenceSteps()

◆ SetLearningRate()

◆ SetTestInterval()

◆ Step() [1/2]

◆ Step() [2/2]

◆ StepLoss()

◆ StepMomentum()

◆ StepNesterov()

◆ StepReducedWeights()

◆ StepReducedWeightsLoss()

Member Data Documentation

◆ fBatchSize

◆ fConvergenceCount

◆ fConvergenceSteps

◆ fLearningRate

◆ fMinimumError

◆ fStepCount

◆ fTestError

◆ fTestInterval

◆ fTrainingError