VectorRandomizer Class Reference

Randomizes elements of a vector according to a mask. More...

#include <nnet-randomizer.h>

Collaboration diagram for VectorRandomizer:

Public Member Functions
	VectorRandomizer ()
	VectorRandomizer (const NnetDataRandomizerOptions &conf)
void	Init (const NnetDataRandomizerOptions &conf)
	Set the randomizer parameters (size) More...
void	AddData (const Vector< BaseFloat > &v)
	Add data to randomization buffer. More...
bool	IsFull ()
	Returns true, when capacity is full. More...
int32	NumFrames ()
	Number of frames stored inside the Randomizer. More...
void	Randomize (const std::vector< int32 > &mask)
	Randomize matrix row-order using mask. More...
bool	Done ()
	Returns true, if no more data for another mini-batch (after current one) More...
void	Next ()
	Sets cursor to next mini-batch. More...
const Vector< BaseFloat > &	Value ()
	Returns matrix-window with next mini-batch. More...

Private Attributes
Vector< BaseFloat >	data_
Vector< BaseFloat >	minibatch_
int32	data_begin_
	A cursor, pointing to the 'row' where the next mini-batch begins,. More...
int32	data_end_
	A cursor, pointing to the 'row' after the end of data,. More...
NnetDataRandomizerOptions	conf_

Detailed Description

Randomizes elements of a vector according to a mask.

Definition at line 148 of file nnet-randomizer.h.

Constructor & Destructor Documentation

VectorRandomizer() [1/2]

VectorRandomizer ( )

inline

Definition at line 150 of file nnet-randomizer.h.

                    :
    data_begin_(0),
    data_end_(0)
  { }

VectorRandomizer() [2/2]

VectorRandomizer ( const NnetDataRandomizerOptions &  conf )

inlineexplicit

Definition at line 155 of file nnet-randomizer.h.

                                                                  :
    data_begin_(0),
    data_end_(0)
  {
    Init(conf);
  }

Member Function Documentation

AddData()

void AddData

(

const Vector< BaseFloat > &

v

)

Add data to randomization buffer.

Definition at line 114 of file nnet-randomizer.cc.

References data_, VectorBase< Real >::Dim(), and KALDI_ASSERT.

Referenced by main(), and UnitTestVectorRandomizer().

                                                         {
  // pre-allocate before 1st use
  if (data_.Dim() == 0) {
    data_.Resize(conf_.randomizer_size);
  }
  // optionally put previous left-over to front
  if (data_begin_ > 0) {
    KALDI_ASSERT(data_begin_ <= data_end_);  // sanity check
    int32 leftover = data_end_ - data_begin_;
    KALDI_ASSERT(leftover < data_begin_);  // no overlap
    if (leftover > 0) {
      data_.Range(0, leftover).CopyFromVec(data_.Range(data_begin_, leftover));
    }
    data_begin_ = 0;
    data_end_ = leftover;
    data_.Range(leftover, data_.Dim()-leftover).SetZero();  // zeroing the rest
  }
  // extend the buffer if necessary
  if (data_.Dim() < data_end_ + v.Dim()) {
    Vector<BaseFloat> data_aux(data_);
    data_.Resize(data_end_ + v.Dim() + 1000);  // +1000 row surplus
    data_.Range(0, data_aux.Dim()).CopyFromVec(data_aux);
  }
  // copy the data
  data_.Range(data_end_, v.Dim()).CopyFromVec(v);
  data_end_ += v.Dim();
}

Done()

bool Done ( )

inline

Returns true, if no more data for another mini-batch (after current one)

Definition at line 184 of file nnet-randomizer.h.

Referenced by UnitTestVectorRandomizer().

              {
    return (data_end_ - data_begin_ < conf_.minibatch_size);
  }

Init()

void Init ( const NnetDataRandomizerOptions &  conf )

inline

Set the randomizer parameters (size)

Definition at line 163 of file nnet-randomizer.h.

Referenced by UnitTestVectorRandomizer().

                                                   {
    conf_ = conf;
  }

IsFull()

bool IsFull ( )

inline

Returns true, when capacity is full.

Definition at line 171 of file nnet-randomizer.h.

Referenced by UnitTestVectorRandomizer().

                {
    return ((data_begin_ == 0) && (data_end_ > conf_.randomizer_size ));
  }

Next()

void Next

(

)

Sets cursor to next mini-batch.

Definition at line 154 of file nnet-randomizer.cc.

Referenced by main(), and UnitTestVectorRandomizer().

                            {
  data_begin_ += conf_.minibatch_size;
}

NumFrames()

int32 NumFrames ( )

inline

Number of frames stored inside the Randomizer.

Definition at line 176 of file nnet-randomizer.h.

References kaldi::cu::Randomize().

Referenced by UnitTestVectorRandomizer().

                    {
    return data_end_;
  }

Randomize()

void Randomize

(

const std::vector< int32 > &

mask

)

Randomize matrix row-order using mask.

Definition at line 142 of file nnet-randomizer.cc.

References data_, rnnlm::i, and KALDI_ASSERT.

Referenced by main(), and UnitTestVectorRandomizer().

                                                             {
  KALDI_ASSERT(data_begin_ == 0);
  KALDI_ASSERT(data_end_ > 0);
  KALDI_ASSERT(data_end_ == mask.size());
  // Use auxiliary buffer for unshuffled data
  Vector<BaseFloat> data_aux(data_);
  // randomize the data, mask is used to index elements in source vector
  for (int32 i = 0; i < mask.size(); i++) {
    data_(i) = data_aux(mask.at(i));
  }
}

Value()

const Vector< BaseFloat > & Value

(

)

Returns matrix-window with next mini-batch.

Definition at line 158 of file nnet-randomizer.cc.

References data_, KALDI_ASSERT, and kaldi::kUndefined.

Referenced by main(), and UnitTestVectorRandomizer().

                                                 {
  // make sure we have data for next minibatch,
  KALDI_ASSERT(data_end_ - data_begin_ >= conf_.minibatch_size);
  // prepare the mini-batch buffer,
  minibatch_.Resize(conf_.minibatch_size, kUndefined);
  minibatch_.CopyFromVec(data_.Range(data_begin_, conf_.minibatch_size));
  return minibatch_;
}

Member Data Documentation

conf_

NnetDataRandomizerOptions conf_

private

Definition at line 203 of file nnet-randomizer.h.

data_

Vector<BaseFloat> data_

private

Definition at line 195 of file nnet-randomizer.h.

data_begin_

int32 data_begin_

private

A cursor, pointing to the 'row' where the next mini-batch begins,.

Definition at line 199 of file nnet-randomizer.h.

data_end_

int32 data_end_

private

A cursor, pointing to the 'row' after the end of data,.

Definition at line 201 of file nnet-randomizer.h.

minibatch_

Vector<BaseFloat> minibatch_

private

Definition at line 196 of file nnet-randomizer.h.

---

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

• nnet/nnet-randomizer.h
• nnet/nnet-randomizer.cc