ceras::model< Ex, Ph > Struct Template Reference

#include <model.hpp>

Public Types
typedef Ph	input_layer_type
typedef Ex	output_layer_type

Public Member Functions
input_layer_type	input () const noexcept
output_layer_type	output () const noexcept
	model (input_layer_type const &place_holder, output_layer_type const &expression)
template<Tensor Tsor>
auto	predict (Tsor const &input_tensor)
template<Expression Exp>
auto	operator() (Exp const &ex) const noexcept
template<typename Loss , typename Optimizer >
auto	compile (Loss const &l, Optimizer const &o)
void	trainable (bool t)
void	save_weights (std::string const &file)
void	load_weights (std::string const &file)
void	summary (std::string const &file_name=std::string{}) const noexcept

Public Attributes
output_layer_type	expression_
	output layer of the model. More...
input_layer_type	place_holder_

Detailed Description

template<Expression Ex, Place_Holder Ph>
struct ceras::model< Ex, Ph >

Groups an input layer (a place holder) and an output layer (an expression template) into an object.

Template Parameters

Ex	The expression template for the output layer.
Ph	The place holder expression for the input layer

Member Typedef Documentation

input_layer_type

template<Expression Ex, Place_Holder Ph>

typedef Ph ceras::model< Ex, Ph >::input_layer_type

output_layer_type

template<Expression Ex, Place_Holder Ph>

typedef Ex ceras::model< Ex, Ph >::output_layer_type

Constructor & Destructor Documentation

model()

template<Expression Ex, Place_Holder Ph>

ceras::model< Ex, Ph >::model ( input_layer_type const &  place_holder, output_layer_type const &  expression  )

inline

Parameters

place_holder	The input layer of the model, a place holder.
expression	The output layer of the model, a expression template.

Example code to generate a model:

auto input = Input();
auto l1 = relu( Dense( 1024, 28*28 )( input ) );
auto output = sigmoid( Dense( 10, 1024 )( l1 ) );
auto m = model{ input, output };

Member Function Documentation

compile()

template<Expression Ex, Place_Holder Ph>

template<typename Loss , typename Optimizer >

auto ceras::model< Ex, Ph >::compile ( Loss const &  l, Optimizer const &  o  )

inline

Compile the model for training

Parameters

l	The loss to minimize.
o	The optimizer to do the optimization.

Returns

An instance of compiled_model.

Example useage:

model m{ ... };
unsigned long batch_size = 16;
float learning_rate = 0.001f;
auto cm = m.compile( MeanSquaredError(), SGD( batch_size, learning_rate ) );

input()

template<Expression Ex, Place_Holder Ph>

input_layer_type ceras::model< Ex, Ph >::input ( ) const

inlinenoexcept

Returns the input layer of the model, which is a place_holder.

load_weights()

template<Expression Ex, Place_Holder Ph>

void ceras::model< Ex, Ph >::load_weights ( std::string const &  file )

inline

Loads all variables from a file

operator()()

template<Expression Ex, Place_Holder Ph>

template<Expression Exp>

auto ceras::model< Ex, Ph >::operator() ( Exp const &  ex ) const

inlinenoexcept

Generating a new expression by using the current model.

Parameters

ex	An expression that represents the input to the model.

Returns

An expression that replacing the input node with a new epxression.

Example code

auto x = Input(); // input, (28*28,)
auto y = Dense( 128, 28*28 )( x );
auto m1 = model( x, y ); // this model is [(28*28,) -> (128,)]
 
auto u = Input(); // new input, (32,)
auto v = Dense( 28*28, 32 )( u );
auto m2 = model( u, v );
 
auto input = Input(); // (32, )
auto ouptut = m1( m2( input ) ); // this new expression is [(32,) -> (28*28,) -> (128,)], note x is not in this expression any more
auto m = model( input, output ); // create a new model

output()

template<Expression Ex, Place_Holder Ph>

output_layer_type ceras::model< Ex, Ph >::output ( ) const

inlinenoexcept

Returns the output layer of the model.

predict()

template<Expression Ex, Place_Holder Ph>

template<Tensor Tsor>

auto ceras::model< Ex, Ph >::predict ( Tsor const &  input_tensor )

inline

Making prediction by binding the nput data to the place_holder_ and evaluating expression_.

Parameters

input_tensor

The input samples.

Returns

The result this model predicts.

Example to predict

auto input = Input();
auto l1 = relu( Dense( 1024, 28*28 )( input ) );
auto output = sigmoid( Dense( 10, 1024 )( l1 ) );
// ... train the model after defining a loss and an optimizer
auto m = model{ input, output };
auto test_data = random( {128, 28*28} ); // batch size is 128
auto result = model.predict( test_data ); // should produce an tensor of (128, 10)

save_weights()

template<Expression Ex, Place_Holder Ph>

void ceras::model< Ex, Ph >::save_weights ( std::string const &  file )

inline

Writes all variables to a file

summary()

template<Expression Ex, Place_Holder Ph>

void ceras::model< Ex, Ph >::summary ( std::string const &  file_name = std::string{} ) const

inlinenoexcept

Print the model summary to console or to a file.

Parameters

file_name

The file to save the summary. If empty, the summary will be printed to console. Empty by default.

trainable()

template<Expression Ex, Place_Holder Ph>

void ceras::model< Ex, Ph >::trainable ( bool  t )

inline

Member Data Documentation

expression_

template<Expression Ex, Place_Holder Ph>

output_layer_type ceras::model< Ex, Ph >::expression_

output layer of the model.

place_holder_

template<Expression Ex, Place_Holder Ph>

input_layer_type ceras::model< Ex, Ph >::place_holder_

---

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

• /home/feng/workspace/github.repo/ceras/include/model.hpp