onert::backend::train::optimizer::SGD Class Reference

SGD optimizer. More...

#include <SGD.h>

Collaboration diagram for onert::backend::train::optimizer::SGD:

Data Structures
struct	Property

Public Types
using	UpdateFactors = exec::train::optimizer::UpdateFactors

Public Member Functions
	SGD ()
	SGD (const Property &props)
	SGD (double lr)
	SGD (const Property &props, double lr)
std::string	name () const override
	Get the name of optimizer.
double	getLearningRate (uint32_t iteration=0) const override
	Get the Learning Rate.
virtual uint32_t	getVarCount () const override
	Get the number of optimizer variables s.
void	applyGradient (const UpdateFactors &factors) const override
	Apply gradient to a trainable tensor.
Public Member Functions inherited from onert::exec::train::optimizer::Optimizer
virtual	~Optimizer ()=default

Detailed Description

SGD optimizer.

Definition at line 35 of file SGD.h.

Member Typedef Documentation

UpdateFactors

using onert::backend::train::optimizer::SGD::UpdateFactors = exec::train::optimizer::UpdateFactors

Definition at line 38 of file SGD.h.

Constructor & Destructor Documentation

SGD() [1/4]

onert::backend::train::optimizer::SGD::SGD ( )

inlineexplicit

Definition at line 48 of file SGD.h.

: _props{}, _learning_rate{0.01} {}

SGD() [2/4]

onert::backend::train::optimizer::SGD::SGD ( const Property &  props )

inlineexplicit

Definition at line 49 of file SGD.h.

: _props{props}, _learning_rate{0.01} {}

SGD() [3/4]

onert::backend::train::optimizer::SGD::SGD ( double  lr )

inlineexplicit

Definition at line 50 of file SGD.h.

: _props{}, _learning_rate{lr} {}

SGD() [4/4]

onert::backend::train::optimizer::SGD::SGD ( const Property &  props, double  lr  )

inlineexplicit

Definition at line 51 of file SGD.h.

: _props{props}, _learning_rate{lr} {}

Member Function Documentation

applyGradient()

void onert::backend::train::optimizer::SGD::applyGradient ( const UpdateFactors &  factors ) const

overridevirtual

Apply gradient to a trainable tensor.

Parameters

factors

UpdateFactors to be used for applying gradient to a trainable tensor

Implements onert::exec::train::optimizer::Optimizer.

Definition at line 38 of file SGD.cc.

{
  auto [grad_tensor, trainable_tensor, training_step] = factors;
  assert(trainable_tensor.data_type() == grad_tensor.data_type());
 
  if (trainable_tensor.getShape() != grad_tensor.getShape())
  {
    throw std::runtime_error("SGD: Invalid gradient tensor");
  }
 
  const auto lr = getLearningRate(training_step);
  switch (grad_tensor.data_type())
  {
    case ir::DataType::FLOAT32:
      nnfw::cker::train::GradientDescent(
        ops::getShape(&trainable_tensor), ops::getBuffer<float>(&trainable_tensor),
        ops::getShape(&grad_tensor), ops::getBuffer<float>(&grad_tensor), lr);
      break;
    default:
      throw std::runtime_error("SGD: Not supported data type");
  }
}

References getLearningRate(), and nnfw::cker::train::GradientDescent().

getLearningRate()

double onert::backend::train::optimizer::SGD::getLearningRate ( uint32_t  iteration = 0 ) const

overridevirtual

Get the Learning Rate.

Parameters

iteration

The number of training steps

Returns

Learning rate

Implements onert::exec::train::optimizer::Optimizer.

Definition at line 32 of file SGD.cc.

{
  // TODO Use iteration, momentum, and nesterov
  return _learning_rate;
}

Referenced by applyGradient().

getVarCount()

virtual uint32_t onert::backend::train::optimizer::SGD::getVarCount ( ) const

inlineoverridevirtual

Get the number of optimizer variables s.

Returns

The number of optimizer variables

Implements onert::exec::train::optimizer::Optimizer.

Definition at line 74 of file SGD.h.

{ return 0; };

name()

std::string onert::backend::train::optimizer::SGD::name ( ) const

inlineoverridevirtual

Get the name of optimizer.

Returns

The name of optimizer

Reimplemented from onert::exec::train::optimizer::Optimizer.

Definition at line 59 of file SGD.h.

{ return std::string{"SGD"}; }

---

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

• runtime/onert/backend/train/optimizer/SGD.h
• runtime/onert/backend/train/optimizer/SGD.cc