Template class representing an nD (n-Dimensional) tensor. Tensor is row-major, i.e. first dimension is height (rows), second is width (cols), third is depth (channels) etc. More...

#include <Matrix.hpp>

Public Member Functions
	Tensor ()

	Tensor (std::initializer_list< size_t > dims_)

	Tensor (std::vector< size_t > dims_)

	Tensor (const Tensor< T > &t)

	Tensor (const mic::types::Matrix< T > &mat_)

const Tensor< T > &	operator= (const Tensor< T > &t)

	~Tensor ()

void	flatten ()

void	conservativeResize (std::vector< size_t > dims_)

void	resize (std::vector< size_t > dims_)

void	elementwiseFunction (T(*func)(T))

void	elementwiseFunctionScalar (T(*func)(T, T), T scalar)

void	normRandReal (float mean=0, float stddev=1)

T *	data ()

std::vector< size_t >	dims ()

size_t	dim (size_t k)

size_t	size ()

void	zeros ()

void	ones ()

void	enumerate ()

void	setValue (T value_)

void	randn (T mean=0, T stddev=1)

void	rand (T min=0, T max=1)

mic::types::Tensor< T >	operator+ (mic::types::Tensor< T > obj_)

mic::types::Tensor< T >	operator- (mic::types::Tensor< T > obj_)

T	sum ()

T &	operator() (size_t index_)

T &	operator() (std::vector< size_t > coordinates_)

const T &	operator() (size_t index_) const

const T &	operator() (std::vector< size_t > coordinates_) const

size_t	getIndex (std::vector< size_t > coordinates_)

Tensor< T >	block (std::vector< std::vector< size_t > > ranges_)

void	concatenate (const Tensor &obj_)

void	concatenate (std::vector< mic::types::Tensor< T > > tensors_)

Private Member Functions
size_t	recursiveIndex (size_t dim_, std::vector< size_t > coordinates_)

void	recursiveBlockCopy (size_t dim_, std::vector< std::vector< size_t > > ranges_, std::vector< size_t > is_, std::vector< size_t > js_, std::vector< size_t > new_dims_, T *tgt_data_ptr_)

size_t	recursiveCalculateTargetIndex (size_t dim_, std::vector< size_t > js_, std::vector< size_t > dims_)

size_t	recursiveCalculateSourceIndex (size_t dim_, std::vector< size_t > is_, size_t offset_)

template<class Archive >
void	save (Archive &ar, const unsigned int version) const

template<class Archive >
void	load (Archive &ar, const unsigned int version)

Private Attributes
size_t	elements

std::vector< size_t >	dimensions

T *	data_ptr

Friends
class	boost::serialization::access

std::ostream &	operator<< (std::ostream &os_, const Tensor &obj_)

Detailed Description

template<typename T>
class mic::types::Tensor< T >

Template class representing an nD (n-Dimensional) tensor. Tensor is row-major, i.e. first dimension is height (rows), second is width (cols), third is depth (channels) etc.

Author: tkornuta

Template Parameters

T	template parameter denoting data type stored in tensor.

Definition at line 49 of file Matrix.hpp.

Constructor & Destructor Documentation

template<typename T>

mic::types::Tensor< T >::Tensor ( )

inline

Default constructor (empty).

Definition at line 69 of file Tensor.hpp.

template<typename T>

mic::types::Tensor< T >::Tensor ( std::initializer_list< size_t > dims_ )

inline

Constructor - sets the tensor dimension and assigns memory.

Parameters

dims_ Tensor dimensions - initilizer list ({ }).

Definition at line 77 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::elements, and mic::types::Tensor< T >::zeros().

Here is the call graph for this function:

template<typename T>

mic::types::Tensor< T >::Tensor ( std::vector< size_t > dims_ )

inline

Constructor - sets the tensor dimension and assigns memory.

Parameters

dims_ Tensor dimensions ({ }, vector<size_t> etc.).

Definition at line 101 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::elements, and mic::types::Tensor< T >::zeros().

Here is the call graph for this function:

template<typename T>

mic::types::Tensor< T >::Tensor ( const Tensor< T > & t )

inline

Copying constructor - copies the values of the given tensor, including tensor dimensions and data.

Parameters

t	The original tensor to be copied.

Definition at line 125 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

mic::types::Tensor< T >::Tensor ( const mic::types::Matrix< T > & mat_ )

inline

Copying constructor - copies the values of the given 2D matrix.

Parameters

t	The original matrix to be copied.

Definition at line 142 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

mic::types::Tensor< T >::~Tensor ( )

inline

Destructor. Frees memory (if it was assigned).

Definition at line 182 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr.

Member Function Documentation

template<typename T>

Tensor<T> mic::types::Tensor< T >::block ( std::vector< std::vector< size_t > > ranges_ )

inline

Returns the (sub)tensor, a new tensor with data being a subtraction from the original tensor taking into account the ranges given as pairs (lower, higher). For dimensions higher than 3 it uses recursive block copy.

Parameters

lower_ Ranges given as pairs (lower, higher) e <0,size-1> for each dimension. If given range consists of a single value, then it is treated as (lower=higher).

Returns: The created subtensor

Definition at line 550 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::recursiveBlockCopy(), and mic::types::Tensor< T >::size().

Referenced by main().

Here is the call graph for this function:

template<typename T>

void mic::types::Tensor< T >::concatenate ( const Tensor< T > & obj_ )

inline

Concatenates two tensors - attaches the tensor passed as argument "to the back". Note: both tensors must have exactly the same dimensions except the 0th dimension.

Parameters

obj_	Tensor to be attached.

Definition at line 617 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

Referenced by main().

template<typename T>

void mic::types::Tensor< T >::concatenate ( std::vector< mic::types::Tensor< T > > tensors_ )

inline

Concatenates a set of tensors - attaches the tensors passed as argument "to the back" of a given tensor. Note: both tensors must have exactly the same dimensions except the 0th dimension.

Parameters

objs_ A list of tensors to be attached.

Definition at line 646 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::conservativeResize ( std::vector< size_t > dims_ )

inline

Resizes the tensor - sets new tensor dimensions. Important note: the total number of elements must remain unchanged.

Parameters

dims_ New dimensions.

Definition at line 200 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

T* mic::types::Tensor< T >::data ( )

inline

Returns pointer to data.

Definition at line 301 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr.

Referenced by mic::types::Matrix< T >::Matrix().

template<typename T>

size_t mic::types::Tensor< T >::dim ( size_t k )

inline

Returns k-th dimension.

Definition at line 315 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions.

template<typename T>

std::vector<size_t> mic::types::Tensor< T >::dims ( )

inline

Returns dimensions.

Definition at line 308 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions.

Referenced by mic::types::Tensor< T >::getIndex(), mic::types::Matrix< T >::Matrix(), mic::types::Tensor< T >::operator+(), and mic::types::Tensor< T >::operator-().

template<typename T>

void mic::types::Tensor< T >::elementwiseFunction ( T(*)(T) func )

inline

Applies the function to all tensor elements.

Parameters

func	The function to be applied. This must be a function with a single argument.

Definition at line 261 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::elementwiseFunctionScalar	(	T(*)(T, T)	func,
		T	scalar
	)

inline

Applies the function to all tensor elements. Additionally passes scalar to the function as second argument.

Parameters

func	The function to be applied. This must be a function with exactly two arguments.
scalar	Scalar passed as second function argument.

Definition at line 273 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::enumerate ( )

inline

Enumerates - sets values of elements to their indices.

Definition at line 346 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

Referenced by main(), and TEST().

template<typename T>

void mic::types::Tensor< T >::flatten ( )

inline

Flattens the tensor - sets dimensions to [ n ].

Definition at line 190 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

Referenced by main().

template<typename T>

size_t mic::types::Tensor< T >::getIndex ( std::vector< size_t > coordinates_ )

inline

Returns the index of an element in nD matrix. For dimensions higher than 3 it uses recursive function.

Parameters

coordinates_ nD vector of element coordinates ({ }, vector<size_t> etc.).

Returns: Index of the element.

Definition at line 520 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::dims(), mic::types::Tensor< T >::recursiveIndex(), and mic::types::Tensor< T >::size().

Referenced by mic::types::Tensor< T >::operator()().

Here is the call graph for this function:

template<typename T>

template<class Archive >

void mic::types::Tensor< T >::load	(	Archive &	ar,
		const unsigned int	version
	)

inlineprivate

Serialization load - loads the tensor object to archive.

Parameters

ar	Used archive.
version	Version of the tensor class (not used currently).

Definition at line 828 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::normRandReal	(	float	mean = `0`,
		float	stddev = `1`
	)

inline

Set values of all matrix elements to random with a normal distribution.

Parameters

mean	Mean
stddev	Variance

Definition at line 286 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::ones ( )

inline

Sets all element values to one.

Definition at line 337 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

T& mic::types::Tensor< T >::operator() ( size_t index_ )

inline

Operator used for setting the value of a given element of nD tensor.

Parameters

index_ Index (a single value).

Returns: The value of the element.

Definition at line 483 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr.

template<typename T>

T& mic::types::Tensor< T >::operator() ( std::vector< size_t > coordinates_ )

inline

Operator used for setting the value of a given element of nD tensor.

Parameters

coordinates_ nD vector of element coordinates ({ }, vector<size_t> etc.).

Returns: The value of the element.

Definition at line 492 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::getIndex().

Here is the call graph for this function:

template<typename T>

const T& mic::types::Tensor< T >::operator() ( size_t index_ ) const

inline

Operator returning the value of a given element of nD tensor.

Parameters

index_ Index (a single value).

Returns: The value of the element.

Definition at line 501 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr.

template<typename T>

const T& mic::types::Tensor< T >::operator() ( std::vector< size_t > coordinates_ ) const

inline

Operator returning the value of a given element of nD tensor.

Parameters

coordinates_ nD vector of element coordinates ({ }, vector<size_t> etc.).

Returns: The value of the element.

Definition at line 510 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::getIndex().

Here is the call graph for this function:

template<typename T>

mic::types::Tensor<T> mic::types::Tensor< T >::operator+ ( mic::types::Tensor< T > obj_ )

inline

Operator returning new tensor being the sum of two tensors.

Parameters

obj_	The second tensor (on the right side of the +).

Returns: New, resulting tensor being the sum.

Definition at line 406 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::dims(), mic::types::Tensor< T >::elements, and mic::types::Tensor< T >::size().

Here is the call graph for this function:

template<typename T>

mic::types::Tensor<T> mic::types::Tensor< T >::operator- ( mic::types::Tensor< T > obj_ )

inline

Operator returning new tensor being the difference of two tensors.

Parameters

obj_	The second tensor (on the right side of the -).

Returns: New, resulting tensor being the tensor difference.

Definition at line 428 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::dims(), mic::types::Tensor< T >::elements, and mic::types::Tensor< T >::size().

Here is the call graph for this function:

template<typename T>

const Tensor<T>& mic::types::Tensor< T >::operator= ( const Tensor< T > & t )

inline

Assign operator - copies the values of the given tensor, including tensor dimensions and data.

Parameters

t	The original tensor to be copied.

Definition at line 158 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::rand	(	T	min = `0`,
		T	max = `1`
	)

inline

Set values of all matrix elements to random numbers from range <min, max> - uniform distribution.

Parameters

min	Min value.
max	Max value.

Returns: Random real value.

Definition at line 387 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::randn	(	T	mean = `0`,
		T	stddev = `1`
	)

inline

Set values of all matrix elements to random with a normal distribution.

Parameters

mean	Mean
stddev	Variance

Definition at line 368 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

Referenced by TEST().

template<typename T>

void mic::types::Tensor< T >::recursiveBlockCopy	(	size_t	dim_,
		std::vector< std::vector< size_t > >	ranges_,
		std::vector< size_t >	is_,
		std::vector< size_t >	js_,
		std::vector< size_t >	new_dims_,
		T *	tgt_data_ptr_
	)

inlineprivate

A recursive function copying the blocks of data from source (current) tensor to the new one. 0th dimension (of the new tensor) is treated as the size of a block. The function calls itself (recursively) in order to generate a vector of indices used subsequently for calculation of positions in both source and target data tables. It uses two helper functions (also recursive) for computation of those positions. WARNING: Tensor dimensions must be > 1!

Parameters

dim_	The currently analyzed dimension of tensor.
ranges_	nD vector of pairs of ranges.
is_	nD vector of indices in the source tensor.
js_	nD vector of indices in the new/target tensor.
new_dims_	nD vector of dimensions of the new tensor.
tgt_data_ptr_	Pointer to the memory block of the new tensor.

Definition at line 726 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::recursiveCalculateSourceIndex(), and mic::types::Tensor< T >::recursiveCalculateTargetIndex().

Referenced by mic::types::Tensor< T >::block().

Here is the call graph for this function:

template<typename T>

size_t mic::types::Tensor< T >::recursiveCalculateSourceIndex	(	size_t	dim_,
		std::vector< size_t >	is_,
		size_t	offset_
	)

inlineprivate

A recursive function responsible for the calculation of position (index) in source (i.e. this) data table.

Parameters

dim_	The currently analyzed dimension of tensor.
is_	nD vector of indices in the source tensor.
offset_	Offset added aat the end(always const = range[0][0]).

Returns: Index in source (this) data table.

Definition at line 788 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions.

Referenced by mic::types::Tensor< T >::recursiveBlockCopy().

template<typename T>

size_t mic::types::Tensor< T >::recursiveCalculateTargetIndex	(	size_t	dim_,
		std::vector< size_t >	js_,
		std::vector< size_t >	dims_
	)

inlineprivate

A recursive function responsible for the calculation of position (index) in target (i.e. new) data table.

Parameters

dim_	The currently analyzed dimension of tensor.
js_	nD vector of indices in the new/target tensor.
dims_	nD vector of dimensions of the (new) tensor.

Returns: Index in target (new) data table.

Definition at line 761 of file Tensor.hpp.

Referenced by mic::types::Tensor< T >::recursiveBlockCopy().

template<typename T>

size_t mic::types::Tensor< T >::recursiveIndex	(	size_t	dim_,
		std::vector< size_t >	coordinates_
	)

inlineprivate

Recursive method computing the index of element in nD matrix.

Parameters

dim_	Dimension considered at the moment.
coordinates_	nD vector of coordinates.

Returns: Index of the element.

Definition at line 705 of file Tensor.hpp.

References mic::types::Tensor< T >::dimensions.

Referenced by mic::types::Tensor< T >::getIndex().

template<typename T>

void mic::types::Tensor< T >::resize ( std::vector< size_t > dims_ )

inline

Resizes the tensor - sets new tensor dimensions. Analogically to Eigen, the resize() method is a no-operation if the actual matrix size doesn't change. Otherwise, i.e. if the total number of elements must change, a new memory block will be allocated and the method will copy as much elements from the old memory block as possible. If you want a conservative variant of resize() which does not change the coefficients, use conservativeResize(), If the new block will be bigger than the old one the values of "new elements" will be set to zeros.

Parameters

dims_ New dimensions.

Definition at line 224 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, mic::types::Tensor< T >::elements, and mic::types::Tensor< T >::zeros().

Here is the call graph for this function:

template<typename T>

template<class Archive >

void mic::types::Tensor< T >::save	(	Archive &	ar,
		const unsigned int	version
	)		const

inlineprivate

Serialization save - saves the tensor object to archive.

Parameters

ar	Used archive.
version	Version of the tensor class (not used currently).

Definition at line 816 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, mic::types::Tensor< T >::dimensions, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::setValue ( T value_ )

inline

Sets values of all element to the value given as parameter.

Parameters

value_ The value to be set.

Definition at line 356 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

size_t mic::types::Tensor< T >::size ( )

inline

Returns size - number of elements.

Returns: Number of elements.

Definition at line 323 of file Tensor.hpp.

References mic::types::Tensor< T >::elements.

Referenced by mic::types::Tensor< T >::block(), mic::types::Tensor< T >::getIndex(), mic::types::Matrix< T >::Matrix(), mic::types::Tensor< T >::operator+(), and mic::types::Tensor< T >::operator-().

template<typename T>

T mic::types::Tensor< T >::sum ( )

inline

Functions sums the tensor elements.

Returns: Sum of tensor elements.

Definition at line 449 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

template<typename T>

void mic::types::Tensor< T >::zeros ( )

inline

Sets all element values to zero.

Definition at line 330 of file Tensor.hpp.

References mic::types::Tensor< T >::data_ptr, and mic::types::Tensor< T >::elements.

Referenced by mic::types::Tensor< T >::resize(), and mic::types::Tensor< T >::Tensor().

Friends And Related Function Documentation

template<typename T>

friend class boost::serialization::access

friend

Definition at line 808 of file Tensor.hpp.

template<typename T>

std::ostream& operator<<	(	std::ostream &	os_,
		const Tensor< T > &	obj_
	)

friend

Stream operator enabling to print tensor dimensions and all values.

Parameters

os_	Ostream object.
obj_	Tensor object.

Definition at line 463 of file Tensor.hpp.

Member Data Documentation

template<typename T>

T* mic::types::Tensor< T >::data_ptr

private

Table of elements of data types.

Definition at line 697 of file Tensor.hpp.

template<typename T>

std::vector<size_t> mic::types::Tensor< T >::dimensions

private

Dimensions.

Definition at line 692 of file Tensor.hpp.

template<typename T>

size_t mic::types::Tensor< T >::elements

private

Number of elements.

Definition at line 687 of file Tensor.hpp.

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

src/types/Matrix.hpp
src/types/Tensor.hpp

Public Member Functions

Private Member Functions

Private Attributes

Friends

Detailed Description

template<typename T> class mic::types::Tensor< T >

Constructor & Destructor Documentation

Member Function Documentation

Friends And Related Function Documentation

Member Data Documentation

template<typename T>
class mic::types::Tensor< T >