pytorch::Conv2d Class Reference

#include <Convolution.hpp>

Inheritance diagram for pytorch::Conv2d:

Public Member Functions
	Conv2d (const conv_params_t &params, const tensor &filters, const tensor &bias)
	Constructs a Conv2d object given parameters, filters, and bias tensors. More...
	Conv2d (const conv_params_t &params, const std::string &filters_filename="", const std::vector< int > &filt_dims={}, const std::string &bias_filename="", const std::vector< int > &bias_dims={}, const std::string &python_home="../scripts")
	Constructs a Conv2d object given the filenames and sizes of the requisite tensors. Also requires convolution parameters like the other constructor. More...
	Conv2d (const Conv2d &other)
	Copy constructor, constructs a Conv2d object that is exactly a copy of the argument. More...
virtual	~Conv2d ()=default
	Destructor - for now trivial, may need to take on some functionality. More...
void	add_filters (const std::string &filters_filename, const std::vector< int > &filt_dims)
	Read in filters from a file given here if it wasn't passed to the constructor. Overwrites current contents of this->filters. More...
void	add_bias (const std::string &bias_filename, const std::vector< int > &bias_dims)
	Read in bias from a file given here if it wasn't passed to the constructor. Overwrites current contents of this->bias. More...
std::vector< tensor >	forward (const std::vector< tensor > &input)
	Forward function, takes data and performs the Conv2d operation using the already-initialized filters and bias tensors. More...
std::vector< tensor >	operator() (const std::vector< tensor > &input)
	Forward function, takes data and performs the Conv2d operation using the already-initialized filters and bias tensors. More...

Private Attributes
tensor	filters
tensor	bias
conv_params_t	params
pycpp::py_object	utils
bool	has_bias = false

Constructor & Destructor Documentation

Conv2d() [1/3]

pytorch::Conv2d::Conv2d ( const conv_params_t &  params, const tensor &  filters, const tensor &  bias  )

inlineexplicit

Constructs a Conv2d object given parameters, filters, and bias tensors.

Parameters

params	The convolution parameters like filter size, stride, and padding.
filters	The trained filter tensors. For those comfortable with Py_Cpp.
bias	The trained bias tensors. For those comfortable with Py_Cpp.

Conv2d() [2/3]

pytorch::Conv2d::Conv2d ( const conv_params_t &  params, const std::string &  filters_filename = "", const std::vector< int > &  filt_dims = {}, const std::string &  bias_filename = "", const std::vector< int > &  bias_dims = {}, const std::string &  python_home = "../scripts"  )

inlineexplicit

Constructs a Conv2d object given the filenames and sizes of the requisite tensors. Also requires convolution parameters like the other constructor.

Parameters

params	The convolution parameters like filter size, stride, and padding.
filters_filename	The file where the filters tensor is saved. Will be loaded with numpy.load(filename).
filt_dims	The dimensions of the filter tensor in pytorch convention - (batch, channels, h, w)
bias_filename	The file where the bias tensor is saved. Will be loaded with numpy.load(filename).
bias_dims	The dimensions of the bias tensor in pytorch convention - (batch, channels, h, w)
python_home	Where the utility scripts are - holds the loading script necessary to load up the tensors.

Conv2d() [3/3]

pytorch::Conv2d::Conv2d ( const Conv2d &  other )

inline

Copy constructor, constructs a Conv2d object that is exactly a copy of the argument.

Parameters

other

Another Conv2d object.

~Conv2d()

virtual pytorch::Conv2d::~Conv2d ( )

virtualdefault

Destructor - for now trivial, may need to take on some functionality.

Member Function Documentation

add_bias()

void pytorch::Conv2d::add_bias ( const std::string &  bias_filename, const std::vector< int > &  bias_dims  )

inline

Read in bias from a file given here if it wasn't passed to the constructor. Overwrites current contents of this->bias.

Parameters

bias_filename	The file where the bias tensor is saved. Will be loaded with numpy.load(filename).
bias_dims	The dimensions of the bias tensor in pytorch convention - (batch, channels, h, w)

add_filters()

void pytorch::Conv2d::add_filters ( const std::string &  filters_filename, const std::vector< int > &  filt_dims  )

inline

Read in filters from a file given here if it wasn't passed to the constructor. Overwrites current contents of this->filters.

Parameters

filters_filename	The file where the filters tensor is saved. Will be loaded with numpy.load(filename).
filt_dims	The dimensions of the filter tensor in pytorch convention - (batch, channels, h, w)

forward()

std::vector<tensor> pytorch::Conv2d::forward ( const std::vector< tensor > &  input )

inlinevirtual

Forward function, takes data and performs the Conv2d operation using the already-initialized filters and bias tensors.

Parameters

input

Input data size (h_in, w_in, Cin, batch)

Returns

Convolved data size (h_out, w_out, Cout, batch)

Implements pytorch::Layer.

operator()()

std::vector<tensor> pytorch::Conv2d::operator() ( const std::vector< tensor > &  input )

inlinevirtual

Forward function, takes data and performs the Conv2d operation using the already-initialized filters and bias tensors.

Parameters

input

Input data size (h_in, w_in, Cin, batch)

Returns

Convolved data size (h_out, w_out, Cout, batch)

Implements pytorch::Layer.

Member Data Documentation

bias

tensor pytorch::Conv2d::bias

private

filters

tensor pytorch::Conv2d::filters

private

has_bias

bool pytorch::Conv2d::has_bias = false

private

params

conv_params_t pytorch::Conv2d::params

private

utils

pycpp::py_object pytorch::Conv2d::utils

private

---

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

• Convolution.hpp