Network class. More...

#include <SimdNeural.hpp>

Public Member Functions
	Network ()
	Creates a new object of Network class. More...

void	Clear ()
	Clears all layers of the neural network.

bool	Empty () const
	Returns true if the neural network is empty.

bool	Add (Layer *layer)
	Adds new Layer to the neural network. More...

const Index &	InputIndex () const
	Gets dimensions of input data. More...

const Index &	OutputIndex () const
	Gets dimensions of output data. More...

template<class Logger >
bool	Train (const Vectors &src, const Labels &dst, const TrainOptions &options, Logger logger)
	Trains the neural network. More...

template<class Logger >
bool	Train (const Vectors &src, const Vectors &dst, const TrainOptions &options, Logger logger)
	Trains the neural network. More...

void	SetThreadNumber (size_t number, bool train=false)
	Sets thread number. More...

SIMD_INLINE const Vector &	Predict (const Vector &x, size_t thread=0, Layer::Method method=Layer::Fast)
	Classifies given sample. More...

bool	Load (const void *data, size_t size, bool train=false)
	Loads the weights of neural network from an external buffer. More...

bool	Load (std::istream &is, bool train=false)
	Loads the weights of neural network from file stream. More...

bool	Load (const std::string &path, bool train=false)
	Loads the weights of neural network from file. More...

bool	Save (void data, size_t size, bool train=false) const
	Saves the weights of neural network into external buffer. More...

bool	Save (std::ostream &os, bool train=false) const
	Saves the weights of neural network to file stream. More...

bool	Save (const std::string &path, bool train=false) const
	Saves the weights of neural network to file. More...

void	Convert (const Labels &src, Vectors &dst) const
	Converts format of classification results. More...

Detailed Description

Network class.

Class Network provides functionality for construction, loading, saving, prediction and training of convolutional neural network.

Constructor & Destructor Documentation

◆ Network()

Network ( )

Creates a new object of Network class.

Creates empty network without any layers.

Member Function Documentation

◆ Add()

bool Add ( Layer * layer )

Adds new Layer to the neural network.

Parameters

[in] layer - a pointer to the new layer. You can add ConvolutionalLayer, MaxPoolingLayer and FullyConnectedLayer.

Returns: a result of addition. If added layer is not compatible with previous layer the result might be negative.

◆ InputIndex()

const Index & InputIndex ( ) const

Gets dimensions of input data.

Returns: a dimensions of input data.

◆ OutputIndex()

const Index & OutputIndex ( ) const

Gets dimensions of output data.

Returns: a dimensions of output data.

◆ Train() [1/2]

bool Train	(	const Vectors &	src,
		const Labels &	dst,
		const TrainOptions &	options,
		Logger	logger
	)

Trains the neural network.

Parameters

[in]	src	- a set of input training samples.
[in]	dst	- a set of classification results.
[in]	options	- an options of training process.
[in]	logger	- a functor to log training process.

Returns: a result of the training.

◆ Train() [2/2]

bool Train	(	const Vectors &	src,
		const Vectors &	dst,
		const TrainOptions &	options,
		Logger	logger
	)

Trains the neural network.

Parameters

[in]	src	- a set of input training samples.
[in]	dst	- a set of classification results.
[in]	options	- an options of training process.
[in]	logger	- a functor to log training process.

Returns: a result of the training.

◆ SetThreadNumber()

void SetThreadNumber	(	size_t	number,
		bool	train = `false`
	)

Sets thread number.

Note: Call this function if you want to call method Predict from dirrerent thread.

Parameters

[in]	number	- a number of threads.
[in]	train	- a train process boolean flag. By default it is equal to False.

◆ Predict()

SIMD_INLINE const Vector & Predict	(	const Vector &	x,
		size_t	thread = `0`,
		Layer::Method	method = `Layer::Fast`
	)

Classifies given sample.

Parameters

[in]	x	- an input sample.
[in]	thread	- a work thread number. By default it is equal to 0.
[in]	method	- a method of prediction. By default it is equal to Layer::Fast.

Returns: a result of classification (vector with predicted probabilities).

◆ Load() [1/3]

bool Load	(	const void *	data,
		size_t	size,
		bool	train = `false`
	)

Loads the weights of neural network from an external buffer.

Note: The network has to be created previously with using of methods Clear/Add.

Parameters

[in]	data	- a pointer to the external buffer.
[in]	size	- a size of the external buffer.
[in]	train	- a boolean flag (True - if we need to load temporary training data, False - otherwise). By default it is equal to False.

Returns: a result of loading.

◆ Load() [2/3]

bool Load	(	std::istream &	is,
		bool	train = `false`
	)

Loads the weights of neural network from file stream.

Note: The network has to be created previously with using of methods Clear/Add.

Parameters

[in]	is	- a input stream.
[in]	train	- a boolean flag (True - if we need to load temporary training data, False - otherwise). By default it is equal to False.

Returns: a result of loading.

◆ Load() [3/3]

bool Load	(	const std::string &	path,
		bool	train = `false`
	)

Loads the weights of neural network from file.

Note: The network has to be created previously with using of methods Clear/Add.

Parameters

[in]	path	- a path to input file.
[in]	train	- a boolean flag (True - if we need to load temporary training data, False - otherwise). By default it is equal to False.

Returns: a result of loading.

◆ Save() [1/3]

bool Save	(	void *	data,
		size_t *	size,
		bool	train = `false`
	)		const

Saves the weights of neural network into external buffer.

Parameters

[out]	data	- a pointer to the external buffer.
[in,out]	size	- a pointer to the size of external buffer. Returns requred buffer size.
[in]	train	- a boolean flag (True - if we need to save temporary training data, False - otherwise). By default it is equal to False.

Returns: a result of saving.

◆ Save() [2/3]

bool Save	(	std::ostream &	os,
		bool	train = `false`
	)		const

Saves the weights of neural network to file stream.

Parameters

[out]	os	- a output stream.
[in]	train	- a boolean flag (True - if we need to save temporary training data, False - otherwise). By default it is equal to False.

Returns: a result of saving.

◆ Save() [3/3]

bool Save	(	const std::string &	path,
		bool	train = `false`
	)		const

Saves the weights of neural network to file.

Parameters

[in]	path	- a path to output file.
[in]	train	- a boolean flag (True - if we need to save temporary training data, False - otherwise). By default it is equal to False.

Returns: a result of saving.

◆ Convert()

void Convert	(	const Labels &	src,
		Vectors &	dst
	)		const

Converts format of classification results.

Parameters

[in]	src	- a set of class indexes.
[out]	dst	- a set of vectors with predicted probabilities.

Simd Library Documentation.

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ Network()

Member Function Documentation

◆ Add()

◆ InputIndex()

◆ OutputIndex()

◆ Train() [1/2]

◆ Train() [2/2]

◆ SetThreadNumber()

◆ Predict()

◆ Load() [1/3]

◆ Load() [2/3]

◆ Load() [3/3]

◆ Save() [1/3]

◆ Save() [2/3]

◆ Save() [3/3]

◆ Convert()