iganet::BatchNorm Class Reference

Batch Normalization as described in the paper. More...

#include </home/runner/work/iganet/iganet/include/iganet/net/activation.hpp>

Inheritance diagram for iganet::BatchNorm:

Public Member Functions
	BatchNorm (torch::Tensor running_mean, torch::Tensor running_var, const torch::Tensor &weight, const torch::Tensor &bias, double eps, double momentum, bool training=false)
	Provides the BatchNorm operation.
	BatchNorm (torch::Tensor running_mean, torch::Tensor running_var, torch::nn::functional::BatchNormFuncOptions options={})
	Provides the BatchNorm operation.
	~BatchNorm () override=default
torch::Tensor	apply (const torch::Tensor &input) const override
	Applies the activation function to the given input.
torch::nn::functional::BatchNormFuncOptions &	options ()
	Returns non-constant reference to options.
const torch::nn::functional::BatchNormFuncOptions &	options () const
	Returns constant reference to options.
void	pretty_print (std::ostream &os) const noexcept override
	Returns a string representation of the activation function.
torch::serialize::InputArchive &	read (torch::serialize::InputArchive &archive, const std::string &key="batch_norm") override
	Reads the activation function from a torch::serialize::InputArchive object.
torch::Tensor &	running_mean ()
	Returns non-constant reference to running mean.
const torch::Tensor &	running_mean () const
	Returns constant reference to running mean.
torch::Tensor &	running_var ()
	Returns non-constant reference to running var.
const torch::Tensor &	running_var () const
	Returns constant reference to running variance.
torch::serialize::OutputArchive &	write (torch::serialize::OutputArchive &archive, const std::string &key="batch_norm") const override
	Writes the activation function into a torch::serialize::OutputArchive object.
Public Member Functions inherited from iganet::ActivationFunction
	~ActivationFunction () override=default

Private Attributes
torch::nn::functional::BatchNormFuncOptions	options_
torch::Tensor	running_mean_
torch::Tensor	running_var_

Additional Inherited Members
Protected Member Functions inherited from iganet::utils::FullQualifiedName
virtual	~FullQualifiedName ()=default
virtual const std::string &	name () const noexcept
	Returns the full qualified name of the object.
Protected Attributes inherited from iganet::utils::FullQualifiedName
at::optional< std::string >	name_
	String storing the full qualified name of the object.

Detailed Description

Batch Normalization as described in the paper.

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift, https://arxiv.org/abs/1502.03167.

Constructor & Destructor Documentation

BatchNorm() [1/2]

iganet::BatchNorm::BatchNorm ( torch::Tensor  running_mean, torch::Tensor  running_var, torch::nn::functional::BatchNormFuncOptions  options = {}  )

inlineexplicit

Provides the BatchNorm operation.

Parameters

running_mean	Value of running_mean.
running_var	Value of running_var.
options	Configuration options.

BatchNorm() [2/2]

iganet::BatchNorm::BatchNorm ( torch::Tensor  running_mean, torch::Tensor  running_var, const torch::Tensor &  weight, const torch::Tensor &  bias, double  eps, double  momentum, bool  training = false  )

inlineexplicit

Provides the BatchNorm operation.

Parameters

running_mean	Value of running_mean.
running_var	Value of running_var.
weight	Value of weight.
bias	Value of bias.
eps	Value of eps.
momentum	Value of momentum.
training	Value of training.

~BatchNorm()

iganet::BatchNorm::~BatchNorm ( )

overridedefault

Member Function Documentation

apply()

torch::Tensor iganet::BatchNorm::apply ( const torch::Tensor &  input ) const

inlineoverridevirtual

Applies the activation function to the given input.

Parameters

input

Tensor to which the activation function is applied.

Returns

The activated tensor.

Implements iganet::ActivationFunction.

options() [1/2]

torch::nn::functional::BatchNormFuncOptions & iganet::BatchNorm::options ( )

inline

Returns non-constant reference to options.

Returns

Mutable reference to the activation options.

options() [2/2]

const torch::nn::functional::BatchNormFuncOptions & iganet::BatchNorm::options ( ) const

inline

Returns constant reference to options.

Returns

Constant reference to the activation options.

pretty_print()

void iganet::BatchNorm::pretty_print ( std::ostream &  os ) const

inlineoverridevirtualnoexcept

Returns a string representation of the activation function.

Parameters

os	Stream that receives the representation.

Implements iganet::ActivationFunction.

read()

torch::serialize::InputArchive & iganet::BatchNorm::read ( torch::serialize::InputArchive &  archive, const std::string &  key = "batch_norm"  )

inlineoverridevirtual

Reads the activation function from a torch::serialize::InputArchive object.

Parameters

archive	Input archive from which the activation state is read.
key	Key prefix under which the state is stored.

Returns

A reference to archive.

Implements iganet::ActivationFunction.

running_mean() [1/2]

torch::Tensor & iganet::BatchNorm::running_mean ( )

inline

Returns non-constant reference to running mean.

Returns

Result of the operation.

running_mean() [2/2]

const torch::Tensor & iganet::BatchNorm::running_mean ( ) const

inline

Returns constant reference to running mean.

Returns

Result of the operation.

running_var() [1/2]

torch::Tensor & iganet::BatchNorm::running_var ( )

inline

Returns non-constant reference to running var.

Returns

Result of the operation.

running_var() [2/2]

const torch::Tensor & iganet::BatchNorm::running_var ( ) const

inline

Returns constant reference to running variance.

Returns

Result of the operation.

write()

torch::serialize::OutputArchive & iganet::BatchNorm::write ( torch::serialize::OutputArchive &  archive, const std::string &  key = "batch_norm"  ) const

inlineoverridevirtual

Writes the activation function into a torch::serialize::OutputArchive object.

Parameters

archive	Output archive that receives the activation state.
key	Key prefix under which the state is stored.

Returns

A reference to archive.

Implements iganet::ActivationFunction.

Member Data Documentation

options_

torch::nn::functional::BatchNormFuncOptions iganet::BatchNorm::options_

private

running_mean_

torch::Tensor iganet::BatchNorm::running_mean_

private

running_var_

torch::Tensor iganet::BatchNorm::running_var_

private

---

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

• include/iganet/net/activation.hpp