mlask::Model Class Reference

Class representing a neural network model. More...

#include <Model.hpp>

Public Member Functions
	Model (std::size_t in, std::size_t out, std::size_t size=0, std::size_t epochs=0, bool log=false)
	Constructs a model.
void	addLayer (std::unique_ptr< Layer > layer)
	Adds a layer to the model(Layer is an abstract class, see Layer.hpp for more details)
template<TLayer layer, typename... Args>
void	addLayer (Args &&... args)
	Adds a layer to the model(Layer is an abstract class, see Layer.hpp for more details)
template<std::size_t in, std::size_t out>
void	addFullyConnectedLayer ()
	creates and adds fullyConnectedLayer with given in and out neurons
void	addLambdaActivationFunction (actfunc func, actfunc derv)
	creates and adds ActivationFunction layer with give function form and derived form
template<ErrorFunction err>
void	backprop (vectorIn input, vectorOut expected)
	performs backpropagation algorithm
void	fit (float_t learning_rate)
	Fits the entire model(every layer added)
vectorOut	forward (vectorIn input) const
	calculates the output for given input
template<ErrorFunction err>
float_t	error (vectorIn input, vectorOut expected) const
	calculates an error of a model
template<ErrorFunction err>
float_t	whole_error (Eigen::Matrix< float_t, Eigen::Dynamic, Eigen::Dynamic > input, Eigen::Matrix< float_t, Eigen::Dynamic, Eigen::Dynamic > expected) const
	calculates an error of a model based on many inputs
const Layer *	getLayer (std::size_t index) const
	gets the layer at a given index
const Layer *	operator[] (std::size_t index) const
std::string	str () const
	Returns a string representation of the model.
void	exportToONNX (std::filesystem::path path, std::string name="MLask Model") const
	exports the model to a file in ONNX format

Detailed Description

Class representing a neural network model.

Constructor & Destructor Documentation

Model()

mlask::Model::Model	(	std::size_t	in,
		std::size_t	out,
		std::size_t	size = 0,
		std::size_t	epochs = 0,
		bool	log = false
	)

Constructs a model.

Parameters

in	number of input neurons
out	number of output neurons
size	number of layers in the model (optional for optimization)
epochs	number of epochs to train the model (optional for logging)
log	whether to log the training process (optional for logging)

Member Function Documentation

addFullyConnectedLayer()

template<std::size_t in, std::size_t out>

void mlask::Model::addFullyConnectedLayer

(

)

creates and adds fullyConnectedLayer with given in and out neurons

Template Parameters

in	number of input neurons
out	number of output neurons

Exceptions

ArchitectureError

if the input size of the layer does not match the output size of the previous layer

addLambdaActivationFunction()

void mlask::Model::addLambdaActivationFunction	(	actfunc	func,
		actfunc	derv
	)

creates and adds ActivationFunction layer with give function form and derived form

Parameters

func	The activation function
derv	The derivative of the activation function

addLayer() [1/2]

template<TLayer layer, typename... Args>

void mlask::Model::addLayer

(

Args &&...

args

)

Adds a layer to the model(Layer is an abstract class, see Layer.hpp for more details)

Template Parameters

layer

The layer to add, a class needs to derive from layer abstract class, see layer.hpp and TLayer concept

This one doesn't check if in and out are valid

addLayer() [2/2]

void mlask::Model::addLayer

(

std::unique_ptr< Layer >

layer

)

Adds a layer to the model(Layer is an abstract class, see Layer.hpp for more details)

Parameters

layer

The layer to add

Exceptions

ArchitectureError

if the input size of the layer does not match the output size of the previous layer

backprop()

template<ErrorFunction err>

void mlask::Model::backprop	(	vectorIn	input,
		vectorOut	expected
	)

performs backpropagation algorithm

Parameters

input	the input vector for the model
expected	the expected output vector for the given input

Template Parameters

err	ErrorFunction used to calculate error, notice that this should be derived version

error()

template<ErrorFunction err>

float_t mlask::Model::error	(	vectorIn	input,
		vectorOut	expected
	)		const

calculates an error of a model

Parameters

input	The input vector for the model
expected	The expected vector of values

Template Parameters

err	the error function (see ErrorFunction.hpp or include/error_functions)

Returns

the error based on an error function

exportToONNX()

void mlask::Model::exportToONNX	(	std::filesystem::path	path,
		std::string	name = "MLask Model"
	)		const

exports the model to a file in ONNX format

Parameters

path	The path to the file where the model will be saved
name	The name of the model to be used in the ONNX file (optional, default is "MLask Model")

Exceptions

throws

ExportError if the export failed

fit()

void mlask::Model::fit

(

float_t

learning_rate

)

Fits the entire model(every layer added)

Parameters

learning_rate

The learning rate for the model

forward()

vectorOut mlask::Model::forward

(

vectorIn

input

)

const

calculates the output for given input

Parameters

input

The input vector for the model

Returns

The output vector for the given input

getLayer()

const Layer * mlask::Model::getLayer ( std::size_t  index ) const

inline

gets the layer at a given index

Parameters

index

The index of the layer to get

Returns

A pointer to the layer at the given index

operator[]()

const Layer * mlask::Model::operator[] ( std::size_t  index ) const

inline

str()

std::string mlask::Model::str

(

)

const

Returns a string representation of the model.

whole_error()

template<ErrorFunction err>

float_t mlask::Model::whole_error	(	Eigen::Matrix< float_t, Eigen::Dynamic, Eigen::Dynamic >	input,
		Eigen::Matrix< float_t, Eigen::Dynamic, Eigen::Dynamic >	expected
	)		const

calculates an error of a model based on many inputs

Parameters

input	The input matrix for the model
expected	The expected matrix of values

Template Parameters

err	the error function (see ErrorFunction.hpp or include/error_functions)

Returns

the error based on an error function

---

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

• include/Model.hpp
• src/Model.cpp