sopt::wavelets::SARA Class Reference

Sparsity Averaging Reweighted Analysis. More...

#include <sara.h>

Inheritance diagram for sopt::wavelets::SARA:

Collaboration diagram for sopt::wavelets::SARA:

Public Member Functions
	SARA (std::initializer_list< std::tuple< std::string, t_uint >> const &init)
	Easy constructor. More...
template<typename ITERATOR , class T = typename std::enable_if< std::is_convertible<decltype(std::get<0>(*std::declval<ITERATOR>())), std::string>::value and std::is_convertible<decltype(std::get<1>(*std::declval<ITERATOR>())), t_uint>::value, ::type >
	SARA (ITERATOR first, ITERATOR last)
	Construct from any iterator over a (std:string, t_uint) tuple. More...
	SARA (const_iterator first, const_iterator last)
virtual	~SARA ()
	Destructor. More...
template<typename T0 >
T0::PlainObject	direct (Eigen::ArrayBase< T0 > const &signal) const
	Direct transform. More...
template<typename T0 , typename T1 >
void	direct (Eigen::ArrayBase< T1 > &coefficients, Eigen::ArrayBase< T0 > const &signal) const
	Direct transform. More...
template<typename T0 , typename T1 >
void	direct (Eigen::ArrayBase< T1 > &&coefficients, Eigen::ArrayBase< T0 > const &signal) const
	Direct transform. More...
template<typename T0 >
T0::PlainObject	indirect (Eigen::ArrayBase< T0 > const &coeffs) const
	Indirect transform. More...
template<typename T0 , typename T1 >
void	indirect (Eigen::ArrayBase< T1 > const &coefficients, Eigen::ArrayBase< T0 > &signal) const
	Indirect transform. More...
template<typename T0 , typename T1 >
void	indirect (Eigen::ArrayBase< T1 > const &coeffs, Eigen::ArrayBase< T0 > &&signal) const
	Indirect transform. More...
t_uint	max_levels () const
	Number of levels over which to do transform. More...
void	emplace_back (std::string const &name, t_uint nlevels)
	Adds a wavelet of specific type. More...

Detailed Description

Sparsity Averaging Reweighted Analysis.

Definition at line 20 of file sara.h.

Constructor & Destructor Documentation

SARA() [1/3]

sopt::wavelets::SARA::SARA ( std::initializer_list< std::tuple< std::string, t_uint >> const &  init )

inline

Easy constructor.

Definition at line 30 of file sara.h.

      : SARA(init.begin(), init.end()) {}

SARA() [2/3]

template<typename ITERATOR , class T = typename std::enable_if< std::is_convertible<decltype(std::get<0>(*std::declval<ITERATOR>())), std::string>::value and std::is_convertible<decltype(std::get<1>(*std::declval<ITERATOR>())), t_uint>::value, ::type >

sopt::wavelets::SARA::SARA ( ITERATOR  first, ITERATOR  last  )

inline

Construct from any iterator over a (std:string, t_uint) tuple.

Definition at line 39 of file sara.h.

                                      {
    for (; first != last; ++first) emplace_back(std::get<0>(*first), std::get<1>(*first));
  }

References emplace_back().

SARA() [3/3]

sopt::wavelets::SARA::SARA ( const_iterator  first, const_iterator  last  )

inline

Definition at line 43 of file sara.h.

: std::vector<Wavelet>(first, last) {}

~SARA()

virtual sopt::wavelets::SARA::~SARA ( )

inlinevirtual

Destructor.

Definition at line 45 of file sara.h.

{}

Member Function Documentation

direct() [1/3]

template<typename T0 >

T0::PlainObject sopt::wavelets::SARA::direct

(

Eigen::ArrayBase< T0 > const &

signal

)

const

Direct transform.

Parameters

[in]

signal

computes wavelet coefficients for this signal. Its size must be a multiple of $2^l$ where $l$ is the maximum number of levels. Can be a matrix (2d-transform) or a column vector (1-d transform).

Returns

wavelets coefficients arranged by columns: if the input is n by m, then the output is n by m * d, with d the number of wavelets.

Supports 1 and 2 dimensional tranforms for real and complex data.

Definition at line 189 of file sara.h.

                                                                          {
  using t_Output = decltype(this->front().direct(signal));
  t_Output result = t_Output::Zero(signal.rows(), signal.cols() * size());
  (*this).direct(result, signal);
  return result;
}

Referenced by direct().

direct() [2/3]

template<typename T0 , typename T1 >

void sopt::wavelets::SARA::direct ( Eigen::ArrayBase< T1 > &&  coefficients, Eigen::ArrayBase< T0 > const &  signal  ) const

inline

Direct transform.

Parameters

[in,out]	coefficients	Output wavelet coefficients. Must be of the type as the input. If the input is n by m, and l is the number of wavelets, then the output should be n by (m * l).
[in]	signal	computes wavelet coefficients for this signal. Its size must be a multiple of $2^l$ where $l$ is the number of levels. Can be a matrix (2d-transform) or a column vector (1-d transform).

Supports 1 and 2 dimensional tranforms for real and complex data. This version allows non-constant Eigen expressions to be passe on without the ugly const_cast of the cannonical approach.

Definition at line 77 of file sara.h.

                                                                                         {
    direct(coefficients, signal);
  }

References direct().

direct() [3/3]

template<typename T0 , typename T1 >

void sopt::wavelets::SARA::direct	(	Eigen::ArrayBase< T1 > &	coefficients,
		Eigen::ArrayBase< T0 > const &	signal
	)		const

Direct transform.

Parameters

[in,out]	coefficients	Output wavelet coefficients. Must be of the type as the input. If the input is n by m, and d is the number of wavelets, then the output should be n by (m * d).
[in]	signal	computes wavelet coefficients for this signal. Its size must be a multiple of $2^l$ where $l$ is the maximum number of levels. Can be a matrix (2d-transform) or a column vector (1-d transform).

Supports 1 and 2 dimensional tranforms for real and complex data.

Definition at line 126 of file sara.h.

                                                                                    {
  SOPT_WAVELET_ERROR_MACRO(signal);
  if (coeffs.rows() != signal.rows() or coeffs.cols() != signal.cols() * static_cast<t_int>(size()))
    coeffs.derived().resize(signal.rows(), signal.cols() * size());
  if (coeffs.rows() != signal.rows() or coeffs.cols() != signal.cols() * static_cast<t_int>(size()))
    throw std::length_error("Incorrect size for output matrix(or could not resize)");
  if (size() == 0) return;
  auto const Ncols = signal.cols();
  {
#ifndef SOPT_OPENMP
    SOPT_TRACE("Calling direct sara without threads");
#endif
    if (Ncols == 1)
      for (size_type i = 0; i < size(); ++i) at(i).direct(coeffs.col(i), signal.col(0));
    else
      for (size_type i = 0; i < size(); ++i)
        at(i).direct(coeffs.leftCols((i + 1) * Ncols).rightCols(Ncols), signal);
  }
 
  coeffs /= std::sqrt(size());
}

References SOPT_TRACE, and SOPT_WAVELET_ERROR_MACRO.

emplace_back()

void sopt::wavelets::SARA::emplace_back ( std::string const &  name, t_uint  nlevels  )

inline

Adds a wavelet of specific type.

Definition at line 114 of file sara.h.

                                                           {
    std::vector<Wavelet>::emplace_back(factory(name, nlevels));
  }

References sopt::wavelets::factory().

Referenced by main(), and SARA().

indirect() [1/3]

template<typename T0 >

T0::PlainObject sopt::wavelets::SARA::indirect

(

Eigen::ArrayBase< T0 > const &

coeffs

)

const

Indirect transform.

Parameters

[in]

coefficients

Input wavelet coefficients. Its size must be a multiple of $2^l$ where $l$ is the number of levels. Can be a matrix (2d-transform) or a column vector (1-d transform).

Supports 1 and 2 dimensional tranforms for real and complex data.

Definition at line 181 of file sara.h.

                                                                            {
  using t_Output = decltype(this->front().indirect(coeffs));
  t_Output signal = t_Output::Zero(coeffs.rows(), coeffs.cols() / size());
  (*this).indirect(coeffs, signal);
  return signal;
}

Referenced by indirect().

indirect() [2/3]

template<typename T0 , typename T1 >

void sopt::wavelets::SARA::indirect	(	Eigen::ArrayBase< T1 > const &	coefficients,
		Eigen::ArrayBase< T0 > &	signal
	)		const

Indirect transform.

Parameters

[in]	coefficients	Input wavelet coefficients. Its size must be a multiple of $2^l$ where $l$ is the number of levels. Can be a matrix (2d-transform) or a column vector (1-d
[in,out]	signal	Reconstructed signal. Must be of the same size and type as the input.

Supports 1 and 2 dimensional tranforms for real and complex data.

Definition at line 149 of file sara.h.

                                                                                      {
  if (size() == 0) throw std::runtime_error("Empty wavelets: adjoint operation undefined");
  if (signal.cols() == 1) {
    if (coeffs.rows() % (1u << max_levels()) != 0)
      throw std::length_error("Inconsistent number of columns and wavelet levels");
  } else
    SOPT_WAVELET_ERROR_MACRO(coeffs);
  if (coeffs.cols() % size() != 0)
    throw std::length_error(
        "Columns of coefficient matrix and number of wavelets are inconsistent");
  if (coeffs.rows() != signal.rows() or coeffs.cols() != signal.cols() * static_cast<t_int>(size()))
    signal.derived().resize(coeffs.rows(), coeffs.cols() / size());
  if (coeffs.rows() != signal.rows() or coeffs.cols() != signal.cols() * static_cast<t_int>(size()))
    throw std::length_error("Incorrect size for output matrix(or could not resize)");
  auto const Ncols = signal.cols();
  {
#ifndef SOPT_OPENMP
    SOPT_TRACE("Calling indirect sara without threads");
#endif
    signal = T0::Zero(signal.rows(), signal.cols());
    if (Ncols == 1)
      for (size_type i = 0; i < size(); ++i) signal.col(0) += at(i).indirect(coeffs.col(i));
    else
      for (size_type i = 0; i < size(); ++i)
        signal += at(i).indirect(coeffs.leftCols((i + 1) * Ncols).rightCols(Ncols));
  }
  signal /= std::sqrt(size());
}

References max_levels(), SOPT_TRACE, and SOPT_WAVELET_ERROR_MACRO.

indirect() [3/3]

template<typename T0 , typename T1 >

void sopt::wavelets::SARA::indirect ( Eigen::ArrayBase< T1 > const &  coeffs, Eigen::ArrayBase< T0 > &&  signal  ) const

inline

Indirect transform.

Parameters

[in]	coefficients	Input wavelet coefficients. Its size must be a multiple of $2^l$ where $l$ is the number of levels. Can be a matrix (2d-transform) or a column vector (1-d
[in,out]	signal	Reconstructed signal. Must be of the same size and type as the input.

Supports 1 and 2 dimensional tranforms for real and complex data. This version allows non-constant Eigen expressions to be passe on without the ugly const_cast of the cannonical approach.

Definition at line 102 of file sara.h.

                                                                                     {
    indirect(coeffs, signal);
  }

References indirect().

max_levels()

t_uint sopt::wavelets::SARA::max_levels ( ) const

inline

Number of levels over which to do transform.

Definition at line 107 of file sara.h.

                            {
    if (size() == 0) return 0;
    auto cmp = [](Wavelet const &a, Wavelet const &b) { return a.levels() < b.levels(); };
    return std::max_element(begin(), end(), cmp)->levels();
  }

References a, and b.

Referenced by indirect().

---

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

• /home/runner/work/sopt/sopt/cpp/sopt/wavelets/sara.h