Logo ROOT   6.21/01
Reference Guide
TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t > Class Template Reference

template<typename Architecture_t, typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
class TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >

Adam Optimizer class.

This class represents the Adam Optimizer.

Definition at line 44 of file Adam.h.

Public Types

using Matrix_t = typename Architecture_t::Matrix_t
 
using Scalar_t = typename Architecture_t::Scalar_t
 
- Public Types inherited from TMVA::DNN::VOptimizer< Architecture_t, Layer_t, DeepNet_t >
using Matrix_t = typename Architecture_t::Matrix_t
 
using Scalar_t = typename Architecture_t::Scalar_t
 

Public Member Functions

 TAdam (DeepNet_t &deepNet, Scalar_t learningRate=0.001, Scalar_t beta1=0.9, Scalar_t beta2=0.999, Scalar_t epsilon=1e-8)
 Constructor. More...
 
 ~TAdam ()=default
 Destructor. More...
 
Scalar_t GetBeta1 () const
 Getters. More...
 
Scalar_t GetBeta2 () const
 
Scalar_t GetEpsilon () const
 
std::vector< std::vector< Matrix_t > > & GetFirstMomentBiases ()
 
std::vector< Matrix_t > & GetFirstMomentBiasesAt (size_t i)
 
std::vector< std::vector< Matrix_t > > & GetFirstMomentWeights ()
 
std::vector< Matrix_t > & GetFirstMomentWeightsAt (size_t i)
 
std::vector< std::vector< Matrix_t > > & GetSecondMomentBiases ()
 
std::vector< Matrix_t > & GetSecondMomentBiasesAt (size_t i)
 
std::vector< std::vector< Matrix_t > > & GetSecondMomentWeights ()
 
std::vector< Matrix_t > & GetSecondMomentWeightsAt (size_t i)
 
- Public Member Functions inherited from TMVA::DNN::VOptimizer< Architecture_t, Layer_t, DeepNet_t >
 VOptimizer (Scalar_t learningRate, DeepNet_t &deepNet)
 Constructor. More...
 
virtual ~VOptimizer ()=default
 Virtual Destructor. More...
 
size_t GetGlobalStep () const
 
Layer_t * GetLayerAt (size_t i)
 
std::vector< Layer_t * > & GetLayers ()
 
Scalar_t GetLearningRate () const
 Getters. More...
 
void IncrementGlobalStep ()
 Increments the global step. More...
 
void SetLearningRate (size_t learningRate)
 Setters. More...
 
void Step ()
 Performs one step of optimization. More...
 

Protected Member Functions

void UpdateBiases (size_t layerIndex, std::vector< Matrix_t > &biases, const std::vector< Matrix_t > &biasGradients)
 Update the biases, given the current bias gradients. More...
 
void UpdateWeights (size_t layerIndex, std::vector< Matrix_t > &weights, const std::vector< Matrix_t > &weightGradients)
 Update the weights, given the current weight gradients. More...
 

Protected Attributes

Scalar_t fBeta1
 The Beta1 constant used by the optimizer. More...
 
Scalar_t fBeta2
 The Beta2 constant used by the optimizer. More...
 
Scalar_t fEpsilon
 The Smoothing term used to avoid division by zero. More...
 
std::vector< std::vector< Matrix_t > > fFirstMomentBiases
 The decaying average of the first moment of the past bias gradients associated with the deep net. More...
 
std::vector< std::vector< Matrix_t > > fFirstMomentWeights
 The decaying average of the first moment of the past weight gradients associated with the deep net. More...
 
std::vector< std::vector< Matrix_t > > fSecondMomentBiases
 The decaying average of the second moment of the past bias gradients associated with the deep net. More...
 
std::vector< std::vector< Matrix_t > > fSecondMomentWeights
 The decaying average of the second moment of the past weight gradients associated with the deep net. More...
 
- Protected Attributes inherited from TMVA::DNN::VOptimizer< Architecture_t, Layer_t, DeepNet_t >
DeepNet_t & fDeepNet
 The reference to the deep net. More...
 
size_t fGlobalStep
 The current global step count during training. More...
 
Scalar_t fLearningRate
 The learning rate used for training. More...
 

#include <TMVA/DNN/Adam.h>

Inheritance diagram for TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >:
[legend]

Member Typedef Documentation

◆ Matrix_t

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
using TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::Matrix_t = typename Architecture_t::Matrix_t

Definition at line 46 of file Adam.h.

◆ Scalar_t

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
using TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::Scalar_t = typename Architecture_t::Scalar_t

Definition at line 47 of file Adam.h.

Constructor & Destructor Documentation

◆ TAdam()

template<typename Architecture_t , typename Layer_t , typename DeepNet_t >
TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::TAdam ( DeepNet_t &  deepNet,
Scalar_t  learningRate = 0.001,
Scalar_t  beta1 = 0.9,
Scalar_t  beta2 = 0.999,
Scalar_t  epsilon = 1e-8 
)

Constructor.

Definition at line 101 of file Adam.h.

◆ ~TAdam()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::~TAdam ( )
default

Destructor.

Member Function Documentation

◆ GetBeta1()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
Scalar_t TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetBeta1 ( ) const
inline

Getters.

Definition at line 79 of file Adam.h.

◆ GetBeta2()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
Scalar_t TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetBeta2 ( ) const
inline

Definition at line 80 of file Adam.h.

◆ GetEpsilon()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
Scalar_t TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetEpsilon ( ) const
inline

Definition at line 81 of file Adam.h.

◆ GetFirstMomentBiases()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> >& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetFirstMomentBiases ( )
inline

Definition at line 86 of file Adam.h.

◆ GetFirstMomentBiasesAt()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<Matrix_t>& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetFirstMomentBiasesAt ( size_t  i)
inline

Definition at line 87 of file Adam.h.

◆ GetFirstMomentWeights()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> >& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetFirstMomentWeights ( )
inline

Definition at line 83 of file Adam.h.

◆ GetFirstMomentWeightsAt()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<Matrix_t>& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetFirstMomentWeightsAt ( size_t  i)
inline

Definition at line 84 of file Adam.h.

◆ GetSecondMomentBiases()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> >& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetSecondMomentBiases ( )
inline

Definition at line 92 of file Adam.h.

◆ GetSecondMomentBiasesAt()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<Matrix_t>& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetSecondMomentBiasesAt ( size_t  i)
inline

Definition at line 93 of file Adam.h.

◆ GetSecondMomentWeights()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> >& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetSecondMomentWeights ( )
inline

Definition at line 89 of file Adam.h.

◆ GetSecondMomentWeightsAt()

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<Matrix_t>& TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::GetSecondMomentWeightsAt ( size_t  i)
inline

Definition at line 90 of file Adam.h.

◆ UpdateBiases()

template<typename Architecture_t , typename Layer_t , typename DeepNet_t >
auto TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::UpdateBiases ( size_t  layerIndex,
std::vector< Matrix_t > &  biases,
const std::vector< Matrix_t > &  biasGradients 
)
protectedvirtual

Update the biases, given the current bias gradients.

Implements TMVA::DNN::VOptimizer< Architecture_t, Layer_t, DeepNet_t >.

Definition at line 164 of file Adam.h.

◆ UpdateWeights()

template<typename Architecture_t , typename Layer_t , typename DeepNet_t >
auto TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::UpdateWeights ( size_t  layerIndex,
std::vector< Matrix_t > &  weights,
const std::vector< Matrix_t > &  weightGradients 
)
protectedvirtual

Update the weights, given the current weight gradients.

Implements TMVA::DNN::VOptimizer< Architecture_t, Layer_t, DeepNet_t >.

Definition at line 140 of file Adam.h.

Member Data Documentation

◆ fBeta1

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
Scalar_t TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fBeta1
protected

The Beta1 constant used by the optimizer.

Definition at line 50 of file Adam.h.

◆ fBeta2

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
Scalar_t TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fBeta2
protected

The Beta2 constant used by the optimizer.

Definition at line 51 of file Adam.h.

◆ fEpsilon

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
Scalar_t TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fEpsilon
protected

The Smoothing term used to avoid division by zero.

Definition at line 52 of file Adam.h.

◆ fFirstMomentBiases

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> > TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fFirstMomentBiases
protected

The decaying average of the first moment of the past bias gradients associated with the deep net.

Definition at line 56 of file Adam.h.

◆ fFirstMomentWeights

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> > TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fFirstMomentWeights
protected

The decaying average of the first moment of the past weight gradients associated with the deep net.

Definition at line 54 of file Adam.h.

◆ fSecondMomentBiases

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> > TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fSecondMomentBiases
protected

The decaying average of the second moment of the past bias gradients associated with the deep net.

Definition at line 61 of file Adam.h.

◆ fSecondMomentWeights

template<typename Architecture_t , typename Layer_t = VGeneralLayer<Architecture_t>, typename DeepNet_t = TDeepNet<Architecture_t, Layer_t>>
std::vector<std::vector<Matrix_t> > TMVA::DNN::TAdam< Architecture_t, Layer_t, DeepNet_t >::fSecondMomentWeights
protected

The decaying average of the second moment of the past weight gradients associated with the deep net.

Definition at line 59 of file Adam.h.


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