Component Class Reference

Abstract class, basic element of the network, it is a box with defined inputs, outputs, and tranformation functions interface. More...

#include <nnet-component.h>

Inheritance diagram for Component:

Collaboration diagram for Component:

Public Member Functions
	Component ()
virtual std::string	Type () const =0
virtual int32	Index () const
	Returns the index in the sequence of layers in the neural net; intended only to be used in debugging information. More...
virtual void	SetIndex (int32 index)
virtual void	InitFromString (std::string args)=0
	Initialize, typically from a line of a config file. More...
virtual int32	InputDim () const =0
	Get size of input vectors. More...
virtual int32	OutputDim () const =0
	Get size of output vectors. More...
virtual std::vector< int32 >	Context () const
	Return a vector describing the temporal context this component requires for each frame of output, as a sorted list. More...
virtual void	Propagate (const ChunkInfo &in_info, const ChunkInfo &out_info, const CuMatrixBase< BaseFloat > &in, CuMatrixBase< BaseFloat > *out) const =0
	Perform forward pass propagation Input->Output. More...
void	Propagate (const ChunkInfo &in_info, const ChunkInfo &out_info, const CuMatrixBase< BaseFloat > &in, CuMatrix< BaseFloat > *out) const
	A non-virtual propagate function that first resizes output if necessary. More...
virtual void	Backprop (const ChunkInfo &in_info, const ChunkInfo &out_info, const CuMatrixBase< BaseFloat > &in_value, const CuMatrixBase< BaseFloat > &out_value, const CuMatrixBase< BaseFloat > &out_deriv, Component *to_update, CuMatrix< BaseFloat > *in_deriv) const =0
	Perform backward pass propagation of the derivative, and also either update the model (if to_update == this) or update another model or compute the model derivative (otherwise). More...
virtual bool	BackpropNeedsInput () const
virtual bool	BackpropNeedsOutput () const
virtual Component *	Copy () const =0
	Copy component (deep copy). More...
virtual void	Read (std::istream &is, bool binary)=0
virtual void	Write (std::ostream &os, bool binary) const =0
	Write component to stream. More...
virtual std::string	Info () const
virtual	~Component ()

Static Public Member Functions
static Component *	ReadNew (std::istream &is, bool binary)
	Read component from stream. More...
static Component *	NewFromString (const std::string &initializer_line)
	Initialize the Component from one line that will contain first the type, e.g. More...
static Component *	NewComponentOfType (const std::string &type)
	Return a new Component of the given type e.g. More...

Private Member Functions
	KALDI_DISALLOW_COPY_AND_ASSIGN (Component)

Private Attributes
int32	index_

Detailed Description

Abstract class, basic element of the network, it is a box with defined inputs, outputs, and tranformation functions interface.

It is able to propagate and backpropagate exact implementation is to be implemented in descendants.

Definition at line 157 of file nnet-component.h.

Constructor & Destructor Documentation

Component()

Component ( )

inline

Definition at line 159 of file nnet-component.h.

: index_(-1) { }

~Component()

virtual ~Component ( )

inlinevirtual

Definition at line 264 of file nnet-component.h.

{ }

Member Function Documentation

Backprop()

virtual void Backprop ( const ChunkInfo &  in_info, const ChunkInfo &  out_info, const CuMatrixBase< BaseFloat > &  in_value, const CuMatrixBase< BaseFloat > &  out_value, const CuMatrixBase< BaseFloat > &  out_deriv, Component *  to_update, CuMatrix< BaseFloat > *  in_deriv  ) const

pure virtual

Perform backward pass propagation of the derivative, and also either update the model (if to_update == this) or update another model or compute the model derivative (otherwise).

Note: in_value and out_value are the values of the input and output of the component, and these may be dummy variables if respectively BackpropNeedsInput() or BackpropNeedsOutput() return false for that component (not all components need these).

num_chunks lets us treat the input matrix as contiguous-in-time chunks of equal size; it only matters if splicing is involved.

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponent, LogSoftmaxComponent, SoftmaxComponent, ScaleComponent, SoftHingeComponent, RectifiedLinearComponent, PowerComponent, TanhComponent, SigmoidComponent, NormalizeComponent, PnormComponent, MaxpoolingComponent, and MaxoutComponent.

Referenced by NnetComputer::Backprop(), NnetDiscriminativeUpdater::Backprop(), NnetUpdater::Backprop(), NnetRescaler::RescaleComponent(), and kaldi::nnet2::UnitTestGenericComponentInternal().

BackpropNeedsInput()

virtual bool BackpropNeedsInput ( ) const

inlinevirtual

Reimplemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponent, LogSoftmaxComponent, SoftmaxComponent, ScaleComponent, SoftHingeComponent, RectifiedLinearComponent, PowerComponent, TanhComponent, SigmoidComponent, NormalizeComponent, PnormComponent, MaxpoolingComponent, and MaxoutComponent.

Definition at line 235 of file nnet-component.h.

Referenced by NnetComputer::Propagate(), NnetDiscriminativeUpdater::Propagate(), NnetUpdater::Propagate(), and kaldi::nnet2::UnitTestGenericComponentInternal().

{ return true; } // if this returns false,

BackpropNeedsOutput()

virtual bool BackpropNeedsOutput ( ) const

inlinevirtual

Reimplemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponent, LogSoftmaxComponent, SoftmaxComponent, ScaleComponent, SoftHingeComponent, RectifiedLinearComponent, PowerComponent, TanhComponent, SigmoidComponent, NormalizeComponent, PnormComponent, MaxpoolingComponent, and MaxoutComponent.

Definition at line 237 of file nnet-component.h.

References kaldi::cu::Copy().

Referenced by NnetComputer::Propagate(), NnetDiscriminativeUpdater::Propagate(), NnetUpdater::Propagate(), and kaldi::nnet2::UnitTestGenericComponentInternal().

{ return true; } // if this returns false,

Context()

virtual std::vector<int32> Context ( ) const

inlinevirtual

Return a vector describing the temporal context this component requires for each frame of output, as a sorted list.

The default implementation returns a vector ( 0 ), but a splicing layer might return e.g. (-2, -1, 0, 1, 2), but it doesn't have to be contiguous. Note : The context needed by the entire network is a function of the contexts needed by all the components. It is required that Context().front() <= 0 and Context().back() >= 0.

Reimplemented in SpliceMaxComponent, and SpliceComponent.

Definition at line 188 of file nnet-component.h.

Referenced by Nnet::ComputeChunkInfo(), and NnetOnlineComputer::Propagate().

{ return std::vector<int32>(1, 0); }

Copy()

virtual Component* Copy ( ) const

pure virtual

Copy component (deep copy).

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponentPreconditioned, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponentPreconditionedOnline, AffineComponentPreconditioned, AffineComponent, LogSoftmaxComponent, SoftmaxComponent, ScaleComponent, SoftHingeComponent, RectifiedLinearComponent, PowerComponent, TanhComponent, SigmoidComponent, NormalizeComponent, PnormComponent, MaxpoolingComponent, and MaxoutComponent.

Referenced by kaldi::nnet2::InsertComponents(), kaldi::nnet2::ReplaceLastComponents(), and kaldi::nnet2::UnitTestGenericComponentInternal().

Index()

virtual int32 Index ( ) const

inlinevirtual

Returns the index in the sequence of layers in the neural net; intended only to be used in debugging information.

Definition at line 166 of file nnet-component.h.

Referenced by AffineComponentPreconditioned::GetScalingFactor(), and AffineComponentPreconditionedOnline::GetScalingFactor().

{ return index_; }

Info()

std::string Info ( ) const

virtual

Reimplemented in Convolutional1dComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, SpliceMaxComponent, SpliceComponent, AffineComponentPreconditionedOnline, AffineComponentPreconditioned, AffineComponent, ScaleComponent, PowerComponent, PnormComponent, MaxpoolingComponent, MaxoutComponent, and UpdatableComponent.

Definition at line 305 of file nnet-component.cc.

References Component::InputDim(), Component::OutputDim(), and Component::Type().

Referenced by SpliceComponent::Info(), SpliceMaxComponent::Info(), DctComponent::Info(), FixedLinearComponent::Info(), FixedAffineComponent::Info(), FixedScaleComponent::Info(), FixedBiasComponent::Info(), DropoutComponent::Info(), main(), and kaldi::nnet2::UnitTestGenericComponentInternal().

                                {
  std::stringstream stream;
  stream << Type() << ", input-dim=" << InputDim()
         << ", output-dim=" << OutputDim();
  return stream.str();
}

InitFromString()

virtual void InitFromString ( std::string  args )

pure virtual

Initialize, typically from a line of a config file.

The "args" will contain any parameters that need to be passed to the Component, e.g. dimensions.

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponentPreconditioned, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponentPreconditionedOnline, AffineComponentPreconditioned, AffineComponent, ScaleComponent, PowerComponent, PnormComponent, MaxpoolingComponent, MaxoutComponent, and NonlinearComponent.

Referenced by Component::NewFromString().

InputDim()

virtual int32 InputDim ( ) const

pure virtual

Get size of input vectors.

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponent, ScaleComponent, PowerComponent, PnormComponent, MaxpoolingComponent, MaxoutComponent, and NonlinearComponent.

Referenced by Component::Info(), UpdatableComponent::Info(), main(), NnetOnlineComputer::Propagate(), kaldi::nnet2::UnitTestGenericComponentInternal(), and NonlinearComponent::UpdateStats().

KALDI_DISALLOW_COPY_AND_ASSIGN()

KALDI_DISALLOW_COPY_AND_ASSIGN ( Component  )

private

NewComponentOfType()

Component * NewComponentOfType ( const std::string &  type )

static

Return a new Component of the given type e.g.

"SoftmaxComponent", or NULL if no such type exists.

Definition at line 51 of file nnet-component.cc.

Referenced by Component::NewFromString(), and Component::ReadNew().

                                                                        {
  Component *ans = NULL;
  if (component_type == "SigmoidComponent") {
    ans = new SigmoidComponent();
  } else if (component_type == "TanhComponent") {
    ans = new TanhComponent();
  } else if (component_type == "PowerComponent") {
    ans = new PowerComponent();
  } else if (component_type == "SoftmaxComponent") {
    ans = new SoftmaxComponent();
  } else if (component_type == "LogSoftmaxComponent") {
    ans = new LogSoftmaxComponent();
  } else if (component_type == "RectifiedLinearComponent") {
    ans = new RectifiedLinearComponent();
  } else if (component_type == "NormalizeComponent") {
    ans = new NormalizeComponent();
  } else if (component_type == "SoftHingeComponent") {
    ans = new SoftHingeComponent();
  } else if (component_type == "PnormComponent") {
    ans = new PnormComponent();
  } else if (component_type == "MaxoutComponent") {
    ans = new MaxoutComponent();
  } else if (component_type == "ScaleComponent") {
    ans = new ScaleComponent();
  } else if (component_type == "AffineComponent") {
    ans = new AffineComponent();
  } else if (component_type == "AffineComponentPreconditioned") {
    ans = new AffineComponentPreconditioned();
  } else if (component_type == "AffineComponentPreconditionedOnline") {
    ans = new AffineComponentPreconditionedOnline();
  } else if (component_type == "SumGroupComponent") {
    ans = new SumGroupComponent();
  } else if (component_type == "BlockAffineComponent") {
    ans = new BlockAffineComponent();
  } else if (component_type == "BlockAffineComponentPreconditioned") {
    ans = new BlockAffineComponentPreconditioned();
  } else if (component_type == "PermuteComponent") {
    ans = new PermuteComponent();
  } else if (component_type == "DctComponent") {
    ans = new DctComponent();
  } else if (component_type == "FixedLinearComponent") {
    ans = new FixedLinearComponent();
  } else if (component_type == "FixedAffineComponent") {
    ans = new FixedAffineComponent();
  } else if (component_type == "FixedScaleComponent") {
    ans = new FixedScaleComponent();
  } else if (component_type == "FixedBiasComponent") {
    ans = new FixedBiasComponent();
  } else if (component_type == "SpliceComponent") {
    ans = new SpliceComponent();
  } else if (component_type == "SpliceMaxComponent") {
    ans = new SpliceMaxComponent();
  } else if (component_type == "DropoutComponent") {
    ans = new DropoutComponent();
  } else if (component_type == "AdditiveNoiseComponent") {
    ans = new AdditiveNoiseComponent();
  } else if (component_type == "Convolutional1dComponent") {
    ans = new Convolutional1dComponent();
  } else if (component_type == "MaxpoolingComponent") {
    ans = new MaxpoolingComponent();
  }
  return ans;
}

NewFromString()

Component * NewFromString ( const std::string &  initializer_line )

static

Initialize the Component from one line that will contain first the type, e.g.

SigmoidComponent, and then a number of tokens (typically integers or floats) that will be used to initialize the component.

Definition at line 116 of file nnet-component.cc.

References Component::InitFromString(), KALDI_ERR, and Component::NewComponentOfType().

Referenced by Nnet::Init().

                                                                     {
  std::istringstream istr(initializer_line);
  std::string component_type; // e.g. "SigmoidComponent".
  istr >> component_type >> std::ws;
  std::string rest_of_line;
  getline(istr, rest_of_line);
  Component *ans = NewComponentOfType(component_type);
  if (ans == NULL)
    KALDI_ERR << "Bad initializer line (no such type of Component): "
              << initializer_line;
  ans->InitFromString(rest_of_line);
  return ans;
}

OutputDim()

virtual int32 OutputDim ( ) const

pure virtual

Get size of output vectors.

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponent, ScaleComponent, PowerComponent, PnormComponent, MaxpoolingComponent, MaxoutComponent, and NonlinearComponent.

Referenced by kaldi::nnet2::GiveNnetCorrectTopology(), Component::Info(), UpdatableComponent::Info(), and kaldi::nnet2::UnitTestGenericComponentInternal().

Propagate() [1/2]

virtual void Propagate ( const ChunkInfo &  in_info, const ChunkInfo &  out_info, const CuMatrixBase< BaseFloat > &  in, CuMatrixBase< BaseFloat > *  out  ) const

pure virtual

Perform forward pass propagation Input->Output.

Each row is one frame or training example. Interpreted as "num_chunks" equally sized chunks of frames; this only matters for layers that do things like context splicing. Typically this variable will either be 1 (when we're processing a single contiguous chunk of data) or will be the same as in.NumFrames(), but other values are possible if some layers do splicing.

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponent, LogSoftmaxComponent, SoftmaxComponent, ScaleComponent, SoftHingeComponent, RectifiedLinearComponent, PowerComponent, TanhComponent, SigmoidComponent, NormalizeComponent, PnormComponent, MaxpoolingComponent, and MaxoutComponent.

Referenced by NormalizeComponent::BackpropNeedsOutput(), TanhComponent::BackpropNeedsOutput(), RectifiedLinearComponent::BackpropNeedsOutput(), SoftHingeComponent::BackpropNeedsOutput(), ScaleComponent::BackpropNeedsOutput(), SoftmaxComponent::BackpropNeedsOutput(), LogSoftmaxComponent::BackpropNeedsOutput(), AffineComponent::BackpropNeedsOutput(), BlockAffineComponent::BackpropNeedsOutput(), SumGroupComponent::BackpropNeedsOutput(), PermuteComponent::BackpropNeedsOutput(), DropoutComponent::BackpropNeedsOutput(), Convolutional1dComponent::BackpropNeedsOutput(), SpliceComponent::Context(), SpliceMaxComponent::Context(), SigmoidComponent::Copy(), AdditiveNoiseComponent::Copy(), MaxoutComponent::OutputDim(), MaxpoolingComponent::OutputDim(), PnormComponent::OutputDim(), PowerComponent::OutputDim(), DctComponent::OutputDim(), FixedLinearComponent::OutputDim(), FixedAffineComponent::OutputDim(), FixedScaleComponent::OutputDim(), FixedBiasComponent::OutputDim(), NnetComputer::Propagate(), NnetDiscriminativeUpdater::Propagate(), NnetOnlineComputer::Propagate(), NnetUpdater::Propagate(), NnetRescaler::Rescale(), NnetRescaler::RescaleComponent(), and kaldi::nnet2::UnitTestGenericComponentInternal().

Propagate() [2/2]

void Propagate ( const ChunkInfo &  in_info, const ChunkInfo &  out_info, const CuMatrixBase< BaseFloat > &  in, CuMatrix< BaseFloat > *  out  ) const

inline

A non-virtual propagate function that first resizes output if necessary.

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

References ChunkInfo::NumCols(), CuMatrixBase< Real >::NumCols(), ChunkInfo::NumRows(), CuMatrixBase< Real >::NumRows(), and CuMatrix< Real >::Resize().

                                                 {
    if (out->NumRows() != out_info.NumRows() ||
        out->NumCols() != out_info.NumCols()) {
      out->Resize(out_info.NumRows(), out_info.NumCols());
    }

    // Cast to CuMatrixBase to use the virtual version of propagate function.
    Propagate(in_info, out_info, in,
              static_cast<CuMatrixBase<BaseFloat>*>(out));
  }

Read()

virtual void Read ( std::istream &  is, bool  binary  )

pure virtual

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponentPreconditioned, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponentPreconditionedOnline, AffineComponentPreconditioned, AffineComponent, ScaleComponent, PowerComponent, PnormComponent, MaxpoolingComponent, MaxoutComponent, and NonlinearComponent.

Referenced by Component::ReadNew().

ReadNew()

Component * ReadNew ( std::istream &  is, bool  binary  )

static

Read component from stream.

Definition at line 37 of file nnet-component.cc.

References KALDI_ERR, Component::NewComponentOfType(), Component::Read(), and kaldi::ReadToken().

Referenced by Nnet::Read(), and kaldi::nnet2::UnitTestGenericComponentInternal().

                                                         {
  std::string token;
  ReadToken(is, binary, &token); // e.g. "<SigmoidComponent>".
  token.erase(0, 1); // erase "<".
  token.erase(token.length()-1); // erase ">".
  Component *ans = NewComponentOfType(token);
  if (!ans)
    KALDI_ERR << "Unknown component type " << token;
  ans->Read(is, binary);
  return ans;
}

SetIndex()

virtual void SetIndex ( int32  index )

inlinevirtual

Definition at line 168 of file nnet-component.h.

{ index_ = index; }

Type()

virtual std::string Type ( ) const

pure virtual

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponentPreconditioned, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponentPreconditionedOnline, AffineComponentPreconditioned, AffineComponent, LogSoftmaxComponent, SoftmaxComponent, ScaleComponent, SoftHingeComponent, RectifiedLinearComponent, PowerComponent, TanhComponent, SigmoidComponent, NormalizeComponent, PnormComponent, MaxpoolingComponent, and MaxoutComponent.

Referenced by kaldi::nnet2::GiveNnetCorrectTopology(), Component::Info(), UpdatableComponent::Info(), MaxoutComponent::Info(), PnormComponent::Info(), PowerComponent::Info(), ScaleComponent::Info(), NonlinearComponent::InitFromString(), MaxoutComponent::InitFromString(), PnormComponent::InitFromString(), PowerComponent::InitFromString(), NonlinearComponent::Read(), Nnet::ScaleLearningRates(), and NonlinearComponent::Write().

Write()

virtual void Write ( std::ostream &  os, bool  binary  ) const

pure virtual

Write component to stream.

Implemented in Convolutional1dComponent, AdditiveNoiseComponent, DropoutComponent, FixedBiasComponent, FixedScaleComponent, FixedAffineComponent, FixedLinearComponent, DctComponent, PermuteComponent, SumGroupComponent, BlockAffineComponentPreconditioned, BlockAffineComponent, SpliceMaxComponent, SpliceComponent, AffineComponentPreconditionedOnline, AffineComponentPreconditioned, AffineComponent, ScaleComponent, PowerComponent, PnormComponent, MaxpoolingComponent, MaxoutComponent, and NonlinearComponent.

Referenced by kaldi::nnet2::UnitTestGenericComponentInternal().

Member Data Documentation

index_

int32 index_

private

Definition at line 267 of file nnet-component.h.

---

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

• nnet2/nnet-component.h
• nnet2/nnet-component.cc