PldaStats Class Reference

#include <plda.h>

Collaboration diagram for PldaStats:

Classes
struct	ClassInfo

Public Member Functions
	PldaStats ()
void	AddSamples (double weight, const Matrix< double > &group)
	The dimension is set up the first time you add samples. More...
int32	Dim () const
void	Init (int32 dim)
void	Sort ()
bool	IsSorted () const
	~PldaStats ()

Protected Attributes
int32	dim_
int64	num_classes_
int64	num_examples_
double	class_weight_
double	example_weight_
Vector< double >	sum_
SpMatrix< double >	offset_scatter_
std::vector< ClassInfo >	class_info_

Private Member Functions
	KALDI_DISALLOW_COPY_AND_ASSIGN (PldaStats)

Friends
class	PldaEstimator

Detailed Description

Definition at line 172 of file plda.h.

Constructor & Destructor Documentation

PldaStats()

PldaStats ( )

inline

Definition at line 174 of file plda.h.

: dim_(0) { } 

~PldaStats()

~PldaStats

(

)

Definition at line 313 of file plda.cc.

References rnnlm::i.

                      {
  for (size_t i = 0; i < class_info_.size(); i++)
    delete class_info_[i].mean;
}

Member Function Documentation

AddSamples()

void AddSamples	(	double	weight,
		const Matrix< double > &	group
	)

The dimension is set up the first time you add samples.

This function adds training samples corresponding to one class (e.g. a speaker). Each row is a separate sample from this group. The "weight" would normally be 1.0, but you can set it to other values if you want to weight your training samples.

Definition at line 286 of file plda.cc.

References VectorBase< Real >::AddRowSumMat(), KALDI_ASSERT, kaldi::kTrans, rnnlm::n, MatrixBase< Real >::NumCols(), and MatrixBase< Real >::NumRows().

Referenced by main(), and kaldi::UnitTestPldaEstimation().

                                                        {
  if (dim_ == 0) {
    Init(group.NumCols());
  } else {
    KALDI_ASSERT(dim_ == group.NumCols());
  }
  int32 n = group.NumRows(); // number of examples for this class
  Vector<double> *mean = new Vector<double>(dim_);
  mean->AddRowSumMat(1.0 / n, group);

  offset_scatter_.AddMat2(weight, group, kTrans, 1.0);
  // the following statement has the same effect as if we
  // had first subtracted the mean from each element of
  // the group before the statement above.
  offset_scatter_.AddVec2(-n * weight, *mean);

  class_info_.push_back(ClassInfo(weight, mean, n));

  num_classes_ ++;
  num_examples_ += n;
  class_weight_ += weight;
  example_weight_ += weight * n;

  sum_.AddVec(weight, *mean);
}

Dim()

int32 Dim ( ) const

inline

Definition at line 184 of file plda.h.

Referenced by main().

{ return dim_; }

Init()

void Init

(

int32

dim

)

Definition at line 325 of file plda.cc.

References KALDI_ASSERT.

                              {
  KALDI_ASSERT(dim_ == 0);
  dim_ = dim;
  num_classes_ = 0;
  num_examples_ = 0;
  class_weight_ = 0.0;
  example_weight_ = 0.0;
  sum_.Resize(dim);
  offset_scatter_.Resize(dim);
  KALDI_ASSERT(class_info_.empty());
}

IsSorted()

bool IsSorted

(

)

const

Definition at line 318 of file plda.cc.

References rnnlm::i.

Referenced by PldaEstimator::PldaEstimator().

                               {
  for (size_t i = 0; i + 1 < class_info_.size(); i++)
    if (class_info_[i+1] < class_info_[i])
      return false;
  return true;
}

KALDI_DISALLOW_COPY_AND_ASSIGN()

KALDI_DISALLOW_COPY_AND_ASSIGN ( PldaStats  )

private

Sort()

void Sort ( )

inline

Definition at line 188 of file plda.h.

References kaldi::IsSorted().

Referenced by main(), and kaldi::UnitTestPldaEstimation().

{ std::sort(class_info_.begin(), class_info_.end()); }

Friends And Related Function Documentation

PldaEstimator

friend class PldaEstimator

friend

Definition at line 193 of file plda.h.

Member Data Documentation

class_info_

std::vector<ClassInfo> class_info_

protected

Definition at line 220 of file plda.h.

Referenced by PldaEstimator::ComputeObjfPart2(), and PldaEstimator::GetStatsFromClassMeans().

class_weight_

double class_weight_

protected

Definition at line 198 of file plda.h.

Referenced by PldaEstimator::ComputeObjfPart1(), PldaEstimator::ComputeObjfPart2(), PldaEstimator::GetOutput(), PldaEstimator::GetStatsFromClassMeans(), and PldaEstimator::GetStatsFromIntraClass().

dim_

int32 dim_

protected

Definition at line 195 of file plda.h.

example_weight_

double example_weight_

protected

Definition at line 199 of file plda.h.

Referenced by PldaEstimator::ComputeObjf(), PldaEstimator::ComputeObjfPart1(), PldaEstimator::Estimate(), and PldaEstimator::GetStatsFromIntraClass().

num_classes_

int64 num_classes_

protected

Definition at line 196 of file plda.h.

num_examples_

int64 num_examples_

protected

Definition at line 197 of file plda.h.

offset_scatter_

SpMatrix<double> offset_scatter_

protected

Definition at line 204 of file plda.h.

Referenced by PldaEstimator::ComputeObjfPart1(), and PldaEstimator::GetStatsFromIntraClass().

sum_

Vector<double> sum_

protected

Definition at line 201 of file plda.h.

Referenced by PldaEstimator::ComputeObjfPart2(), PldaEstimator::GetOutput(), and PldaEstimator::GetStatsFromClassMeans().

---

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

• ivector/plda.h
• ivector/plda.cc