pytorch::BatchNorm2d Class Reference

#include <Normalization.hpp>

Inheritance diagram for pytorch::BatchNorm2d:

Public Member Functions
	BatchNorm2d (const tensor &gamma, const tensor &beta, const tensor &running_mean, const tensor &running_var, const float &epsilon=1e-5)
	Constructs a BatchNorm2d object. More...
	BatchNorm2d (const std::string &gamma_filename="", const std::vector< int > &gamma_dims={}, const std::string &beta_filename="", const std::vector< int > &beta_dims={}, const std::string &running_mean_filename="", const std::vector< int > &running_mean_dims={}, const std::string &running_var_filename="", const std::vector< int > &running_var_dims={}, const float &epsilon=1e-5, const std::string &python_home="../scripts")
	Constructs a BatchNorm2d object and loads the requisite tensors in from filenames and sizes. More...
void	add_gamma (const std::string &gamma_filename="", const std::vector< int > &gamma_dims={})
	Adds gamma to the layer if the name wasn't passed to the constructor. More...
void	add_beta (const std::string &beta_filename="", const std::vector< int > &beta_dims={})
	Adds beta if it wasn't added by the constructor. More...
void	add_running_mean (const std::string &running_mean_filename="", const std::vector< int > &running_mean_dims={})
	Adds running_mean if it wasn't added by the constructor. More...
void	add_running_var (const std::string &running_var_filename="", const std::vector< int > &running_var_dims={})
	Adds running_var if it wasn't added by the constructor. More...
std::vector< tensor >	forward (const std::vector< tensor > &input)
	Applies the forward pass of batch normalization. More...
std::vector< tensor >	operator() (const std::vector< tensor > &input)
	Applies the forward pass of batch normalization. More...

Private Attributes
tensor	gamma
tensor	beta
tensor	running_mean
tensor	running_var
float	epsilon
pycpp::py_object	utils

Constructor & Destructor Documentation

BatchNorm2d() [1/2]

pytorch::BatchNorm2d::BatchNorm2d ( const tensor &  gamma, const tensor &  beta, const tensor &  running_mean, const tensor &  running_var, const float &  epsilon = 1e-5  )

inline

Constructs a BatchNorm2d object.

Parameters

gamma	The multiplier for the affine transform. Saved as 'bn.weight' by pytorch.
beta	The bias for the affine transform. Saved as 'bn.bias' by pytorch.
running_mean	The running mean for the batchnorm operation.
running_var	The running variance for the batchnorm operation.
epsilon	A factor in the denominator of the transform that adds stability.

BatchNorm2d() [2/2]

pytorch::BatchNorm2d::BatchNorm2d ( const std::string &  gamma_filename = "", const std::vector< int > &  gamma_dims = {}, const std::string &  beta_filename = "", const std::vector< int > &  beta_dims = {}, const std::string &  running_mean_filename = "", const std::vector< int > &  running_mean_dims = {}, const std::string &  running_var_filename = "", const std::vector< int > &  running_var_dims = {}, const float &  epsilon = 1e-5, const std::string &  python_home = "../scripts"  )

inline

Constructs a BatchNorm2d object and loads the requisite tensors in from filenames and sizes.

Parameters

gamma_filename	The file where gamma can be found. Will be loaded with numpy.load(filename).
gamma_dims	The dimensions of gamma in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))
beta_filename	The file where beta can be found. Will be loaded with numpy.load(filename).
beta_dims	The dimensions of beta in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))
running_mean_filename	The file where running_mean can be found. Will be loaded with numpy.load(filename).
running_mean_dims	The dimensions of running_mean in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))
running_var_filename	the file where running_var can be found. Will be loaded with numpy.load(filename).
running_var_dims	The dimensions of running_var in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))
epsilon	A float for numerical stability, 1e-5 by default.
python_home	Where the utility scripts are - holds the loading script necessary to load up the tensors.

Member Function Documentation

add_beta()

void pytorch::BatchNorm2d::add_beta ( const std::string &  beta_filename = "", const std::vector< int > &  beta_dims = {}  )

inline

Adds beta if it wasn't added by the constructor.

Parameters

beta_filename	The file where beta can be found. Will be loaded with numpy.load(filename).
beta_dims	The dimensions of beta in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))

add_gamma()

void pytorch::BatchNorm2d::add_gamma ( const std::string &  gamma_filename = "", const std::vector< int > &  gamma_dims = {}  )

inline

Adds gamma to the layer if the name wasn't passed to the constructor.

Parameters

gamma_filename	The file where gamma can be found. Will be loaded with numpy.load(filename).
gamma_dims	The dimensions of gamma in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))

add_running_mean()

void pytorch::BatchNorm2d::add_running_mean ( const std::string &  running_mean_filename = "", const std::vector< int > &  running_mean_dims = {}  )

inline

Adds running_mean if it wasn't added by the constructor.

Parameters

running_mean_filename	The file where running_mean can be found. Will be loaded with numpy.load(filename).
running_mean_dims	The dimensions of running_mean in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))

add_running_var()

void pytorch::BatchNorm2d::add_running_var ( const std::string &  running_var_filename = "", const std::vector< int > &  running_var_dims = {}  )

inline

Adds running_var if it wasn't added by the constructor.

Parameters

running_var_filename	the file where running_var can be found. Will be loaded with numpy.load(filename).
running_var_dims	The dimensions of running_var in pytorch convention (n, k, h, w) (usually = (1, k, 1, 1))

forward()

std::vector<tensor> pytorch::BatchNorm2d::forward ( const std::vector< tensor > &  input )

inlinevirtual

Applies the forward pass of batch normalization.

Parameters

input

The input data to be normalized.

Returns

The normalized data. The size has not changed.

Implements pytorch::Layer.

operator()()

std::vector<tensor> pytorch::BatchNorm2d::operator() ( const std::vector< tensor > &  input )

inlinevirtual

Applies the forward pass of batch normalization.

Parameters

input

The input data to be normalized.

Returns

The normalized data. The size has not changed.

Implements pytorch::Layer.

Member Data Documentation

beta

tensor pytorch::BatchNorm2d::beta

private

epsilon

float pytorch::BatchNorm2d::epsilon

private

gamma

tensor pytorch::BatchNorm2d::gamma

private

running_mean

tensor pytorch::BatchNorm2d::running_mean

private

running_var

tensor pytorch::BatchNorm2d::running_var

private

utils

pycpp::py_object pytorch::BatchNorm2d::utils

private

---

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

• Normalization.hpp