WARNING, this component is deprecated in favor of TimeHeightConvolutionComponent, and will be deleted. More...

#include <nnet-combined-component.h>

Inheritance diagram for ConvolutionComponent:

Collaboration diagram for ConvolutionComponent:

[legend]

Public Types
enum	TensorVectorizationType { kYzx = 0, kZyx = 1 }

Public Member Functions
	ConvolutionComponent ()

	ConvolutionComponent (const ConvolutionComponent &component)

	ConvolutionComponent (const CuMatrixBase< BaseFloat > &filter_params, const CuVectorBase< BaseFloat > &bias_params, int32 input_x_dim, int32 input_y_dim, int32 input_z_dim, int32 filt_x_dim, int32 filt_y_dim, int32 filt_x_step, int32 filt_y_step, TensorVectorizationType input_vectorization, BaseFloat learning_rate)

virtual int32	InputDim () const
	Returns input-dimension of this component. More...

virtual int32	OutputDim () const
	Returns output-dimension of this component. More...

virtual std::string	Info () const
	Returns some text-form information about this component, for diagnostics. More...

virtual void	InitFromConfig (ConfigLine *cfl)
	Initialize, from a ConfigLine object. More...

virtual std::string	Type () const
	Returns a string such as "SigmoidComponent", describing the type of the object. More...

virtual int32	Properties () const
	Return bitmask of the component's properties. More...

virtual void *	Propagate (const ComponentPrecomputedIndexes indexes, const CuMatrixBase< BaseFloat > &in, CuMatrixBase< BaseFloat > out) const
	Propagate function. More...

virtual void	Backprop (const std::string &debug_info, const ComponentPrecomputedIndexes indexes, const CuMatrixBase< BaseFloat > &in_value, const CuMatrixBase< BaseFloat > &, const CuMatrixBase< BaseFloat > &out_deriv, void memo, Component to_update_in, CuMatrixBase< BaseFloat > in_deriv) const
	Backprop function; depending on which of the arguments 'to_update' and 'in_deriv' are non-NULL, this can compute input-data derivatives and/or perform model update. More...

void	Update (const std::string &debug_info, const CuMatrixBase< BaseFloat > &in_value, const CuMatrixBase< BaseFloat > &out_deriv, const std::vector< CuSubMatrix< BaseFloat > *> &out_deriv_batch)

virtual void	Read (std::istream &is, bool binary)
	Read function (used after we know the type of the Component); accepts input that is missing the token that describes the component type, in case it has already been consumed. More...

virtual void	Write (std::ostream &os, bool binary) const
	Write component to stream. More...

virtual Component *	Copy () const
	Copies component (deep copy). More...

virtual void	Scale (BaseFloat scale)
	This virtual function when called on – an UpdatableComponent scales the parameters by "scale" when called by an UpdatableComponent. More...

virtual void	Add (BaseFloat alpha, const Component &other)
	This virtual function when called by – an UpdatableComponent adds the parameters of another updatable component, times some constant, to the current parameters. More...

virtual void	PerturbParams (BaseFloat stddev)
	This function is to be used in testing. More...

virtual BaseFloat	DotProduct (const UpdatableComponent &other) const
	Computes dot-product between parameters of two instances of a Component. More...

virtual int32	NumParameters () const
	The following new virtual function returns the total dimension of the parameters in this class. More...

virtual void	Vectorize (VectorBase< BaseFloat > *params) const
	Turns the parameters into vector form. More...

virtual void	UnVectorize (const VectorBase< BaseFloat > &params)
	Converts the parameters from vector form. More...

void	SetParams (const VectorBase< BaseFloat > &bias, const MatrixBase< BaseFloat > &filter)

const CuVector< BaseFloat > &	BiasParams () const

const CuMatrix< BaseFloat > &	LinearParams () const

void	Init (int32 input_x_dim, int32 input_y_dim, int32 input_z_dim, int32 filt_x_dim, int32 filt_y_dim, int32 filt_x_step, int32 filt_y_step, int32 num_filters, TensorVectorizationType input_vectorization, BaseFloat param_stddev, BaseFloat bias_stddev)

void	Init (int32 input_x_dim, int32 input_y_dim, int32 input_z_dim, int32 filt_x_dim, int32 filt_y_dim, int32 filt_x_step, int32 filt_y_step, TensorVectorizationType input_vectorization, std::string matrix_filename)

void	Resize (int32 input_dim, int32 output_dim)

void	Update (const std::string &debug_info, const CuMatrixBase< BaseFloat > &in_value, const CuMatrixBase< BaseFloat > &out_deriv)

Public Member Functions inherited from UpdatableComponent
	UpdatableComponent (const UpdatableComponent &other)

	UpdatableComponent ()

virtual	~UpdatableComponent ()

virtual void	SetUnderlyingLearningRate (BaseFloat lrate)
	Sets the learning rate of gradient descent- gets multiplied by learning_rate_factor_. More...

virtual void	SetActualLearningRate (BaseFloat lrate)
	Sets the learning rate directly, bypassing learning_rate_factor_. More...

virtual void	SetAsGradient ()
	Sets is_gradient_ to true and sets learning_rate_ to 1, ignoring learning_rate_factor_. More...

virtual BaseFloat	LearningRateFactor ()

virtual void	SetLearningRateFactor (BaseFloat lrate_factor)

void	SetUpdatableConfigs (const UpdatableComponent &other)

virtual void	FreezeNaturalGradient (bool freeze)
	freezes/unfreezes NaturalGradient updates, if applicable (to be overriden by components that use Natural Gradient). More...

BaseFloat	LearningRate () const
	Gets the learning rate to be used in gradient descent. More...

BaseFloat	MaxChange () const
	Returns the per-component max-change value, which is interpreted as the maximum change (in l2 norm) in parameters that is allowed per minibatch for this component. More...

void	SetMaxChange (BaseFloat max_change)

BaseFloat	L2Regularization () const
	Returns the l2 regularization constant, which may be set in any updatable component (usually from the config file). More...

void	SetL2Regularization (BaseFloat a)

Public Member Functions inherited from Component
virtual void	StoreStats (const CuMatrixBase< BaseFloat > &in_value, const CuMatrixBase< BaseFloat > &out_value, void *memo)
	This function may store stats on average activation values, and for some component types, the average value of the derivative of the nonlinearity. More...

virtual void	ZeroStats ()
	Components that provide an implementation of StoreStats should also provide an implementation of ZeroStats(), to set those stats to zero. More...

virtual void	GetInputIndexes (const MiscComputationInfo &misc_info, const Index &output_index, std::vector< Index > *desired_indexes) const
	This function only does something interesting for non-simple Components. More...

virtual bool	IsComputable (const MiscComputationInfo &misc_info, const Index &output_index, const IndexSet &input_index_set, std::vector< Index > *used_inputs) const
	This function only does something interesting for non-simple Components, and it exists to make it possible to manage optionally-required inputs. More...

virtual void	ReorderIndexes (std::vector< Index > input_indexes, std::vector< Index > output_indexes) const
	This function only does something interesting for non-simple Components. More...

virtual ComponentPrecomputedIndexes *	PrecomputeIndexes (const MiscComputationInfo &misc_info, const std::vector< Index > &input_indexes, const std::vector< Index > &output_indexes, bool need_backprop) const
	This function must return NULL for simple Components. More...

virtual void	DeleteMemo (void *memo) const
	This virtual function only needs to be overwritten by Components that return a non-NULL memo from their Propagate() function. More...

virtual void	ConsolidateMemory ()
	This virtual function relates to memory management, and avoiding fragmentation. More...

	Component ()

virtual	~Component ()

Private Member Functions
void	InputToInputPatches (const CuMatrixBase< BaseFloat > &in, CuMatrix< BaseFloat > *patches) const

void	InderivPatchesToInderiv (const CuMatrix< BaseFloat > &in_deriv_patches, CuMatrixBase< BaseFloat > *in_deriv) const

const ConvolutionComponent &	operator= (const ConvolutionComponent &other)

Private Attributes
int32	input_x_dim_

int32	input_y_dim_

int32	input_z_dim_

int32	filt_x_dim_

int32	filt_y_dim_

int32	filt_x_step_

int32	filt_y_step_

TensorVectorizationType	input_vectorization_

CuMatrix< BaseFloat >	filter_params_

CuVector< BaseFloat >	bias_params_

Additional Inherited Members
Static Public Member Functions inherited from Component
static Component *	ReadNew (std::istream &is, bool binary)
	Read component from stream (works out its type). Dies on error. More...

static Component *	NewComponentOfType (const std::string &type)
	Returns a new Component of the given type e.g. More...

Protected Member Functions inherited from UpdatableComponent
void	InitLearningRatesFromConfig (ConfigLine *cfl)

std::string	ReadUpdatableCommon (std::istream &is, bool binary)

void	WriteUpdatableCommon (std::ostream &is, bool binary) const

Protected Attributes inherited from UpdatableComponent
BaseFloat	learning_rate_
	learning rate (typically 0.0..0.01) More...

BaseFloat	learning_rate_factor_
	learning rate factor (normally 1.0, but can be set to another < value so that when < you call SetLearningRate(), that value will be scaled by this factor. More...

BaseFloat	l2_regularize_
	L2 regularization constant. More...

bool	is_gradient_
	True if this component is to be treated as a gradient rather than as parameters. More...

BaseFloat	max_change_
	configuration value for imposing max-change More...

Detailed Description

WARNING, this component is deprecated in favor of TimeHeightConvolutionComponent, and will be deleted.

ConvolutionalComponent implements 2d-convolution. It uses 3D filters on 3D inputs, but the 3D filters hop only over 2 dimensions as it has same size as the input along the 3rd dimension. Input : A matrix where each row is a vectorized 3D-tensor. The 3D tensor has dimensions x: (e.g. time) y: (e.g. frequency) z: (e.g. channels like features/delta/delta-delta)

The component supports input vectorizations of type zyx and yzx. The default vectorization type is zyx. e.g. for input vectorization of type zyx the input is vectorized by spanning axes z, y and x of the tensor in that order. Given 3d tensor A with sizes (2, 2, 2) along the three dimensions the zyx vectorized input looks like A(0,0,0) A(0,0,1) A(0,1,0) A(0,1,1) A(1,0,0) A(1,0,1) A(1,1,0) A(1,1,1)

Output : The output is also a 3D tensor vectorized in the zyx format. The channel axis (z) in the output corresponds to the output of different filters. The first channel corresponds to the first filter i.e., first row of the filter_params_ matrix.

Note: The component has to support yzx input vectorization as the binaries like add-deltas generate yz vectorized output. These input vectors are concatenated using the Append descriptor across time steps to form a yzx vectorized 3D tensor input. e.g. Append(Offset(input, -1), input, Offset(input, 1))

For information on the hyperparameters and parameters of this component see the variable declarations.

Propagation: ------------ Convolution operation consists of a dot-products between the filter tensor and input tensor patch, for various shifts of filter tensor along the x and y axes input tensor. (Note: there is no shift along z-axis as the filter and input tensor have same size along this axis).

For a particular shift (i,j) of the filter tensor along input tensor dimensions x and y, the elements of the input tensor which overlap with the filter form the input tensor patch. This patch is vectorized in zyx format. All the patches corresponding to various samples in the mini-batch are stacked into a matrix, where each row corresponds to one patch. Let this matrix be represented by X_{i,j}. The dot products with various filters are computed simultaneously by computing the matrix product with the filter_params_ matrix (W) Y_{i,j} = X_{i,j}*W^T. Each row of W corresponds to one filter 3D tensor vectorized in zyx format.

All the matrix products corresponding to various shifts (i,j) of the filter tensor are computed simultaneously using the AddMatMatBatched call of CuMatrixBase class.

BackPropagation: ---------------- Backpropagation to compute the input derivative ( X_{i,j}) consists of the a series of matrix products. {i,j} = {i,j}*W where {i,j} corresponds to the output derivative for a particular shift of the filter.

Once again these matrix products are computed simultaneously.

Update: ------- The weight gradient is computed as = {i,j} (X_{i,j}^T *{i,j})

Definition at line 114 of file nnet-combined-component.h.

Member Enumeration Documentation

◆ TensorVectorizationType

enum TensorVectorizationType

Enumerator
kYzx
kZyx

Definition at line 116 of file nnet-combined-component.h.

                                 {
     kYzx = 0,
     kZyx = 1
   };

Constructor & Destructor Documentation

◆ ConvolutionComponent() [1/3]

ConvolutionComponent ( )

Definition at line 34 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Copy().

                                           :
     UpdatableComponent(),
     input_x_dim_(0), input_y_dim_(0), input_z_dim_(0),
     filt_x_dim_(0), filt_y_dim_(0),
     filt_x_step_(0), filt_y_step_(0),
     input_vectorization_(kZyx) { }

◆ ConvolutionComponent() [2/3]

ConvolutionComponent ( const ConvolutionComponent & component )

Definition at line 41 of file nnet-combined-component.cc.

                                           :
     UpdatableComponent(component),
     input_x_dim_(component.input_x_dim_),
     input_y_dim_(component.input_y_dim_),
     input_z_dim_(component.input_z_dim_),
     filt_x_dim_(component.filt_x_dim_),
     filt_y_dim_(component.filt_y_dim_),
     filt_x_step_(component.filt_x_step_),
     filt_y_step_(component.filt_y_step_),
     input_vectorization_(component.input_vectorization_),
     filter_params_(component.filter_params_),
     bias_params_(component.bias_params_) { }

◆ ConvolutionComponent() [3/3]

ConvolutionComponent	(	const CuMatrixBase< BaseFloat > &	filter_params,
		const CuVectorBase< BaseFloat > &	bias_params,
		int32	input_x_dim,
		int32	input_y_dim,
		int32	input_z_dim,
		int32	filt_x_dim,
		int32	filt_y_dim,
		int32	filt_x_step,
		int32	filt_y_step,
		TensorVectorizationType	input_vectorization,
		BaseFloat	learning_rate
	)

Definition at line 55 of file nnet-combined-component.cc.

References CuVectorBase< Real >::Dim(), UpdatableComponent::is_gradient_, KALDI_ASSERT, CuMatrixBase< Real >::NumCols(), CuMatrixBase< Real >::NumRows(), and UpdatableComponent::SetUnderlyingLearningRate().

                             :
     input_x_dim_(input_x_dim),
     input_y_dim_(input_y_dim),
     input_z_dim_(input_z_dim),
     filt_x_dim_(filt_x_dim),
     filt_y_dim_(filt_y_dim),
     filt_x_step_(filt_x_step),
     filt_y_step_(filt_y_step),
     input_vectorization_(input_vectorization),
     filter_params_(filter_params),
     bias_params_(bias_params){
   KALDI_ASSERT(filter_params.NumRows() == bias_params.Dim() &&
                bias_params.Dim() != 0);
   KALDI_ASSERT(filter_params.NumCols() == filt_x_dim * filt_y_dim * input_z_dim);
   SetUnderlyingLearningRate(learning_rate);
   is_gradient_ = false;
 }

Member Function Documentation

◆ Add()

void Add	(	BaseFloat	alpha,
		const Component &	other
	)

virtual

This virtual function when called by – an UpdatableComponent adds the parameters of another updatable component, times some constant, to the current parameters.

– a NonlinearComponent (or another component that stores stats, like BatchNormComponent)– it relates to adding stats. Otherwise it will normally do nothing.

Reimplemented from Component.

Definition at line 348 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, ConvolutionComponent::filter_params_, and KALDI_ASSERT.

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                                                          {
   const ConvolutionComponent *other =
       dynamic_cast<const ConvolutionComponent*>(&other_in);
   KALDI_ASSERT(other != NULL);
   filter_params_.AddMat(alpha, other->filter_params_);
   bias_params_.AddVec(alpha, other->bias_params_);
 }

◆ Backprop()

void Backprop	(	const std::string &	debug_info,
		const ComponentPrecomputedIndexes *	indexes,
		const CuMatrixBase< BaseFloat > &	in_value,
		const CuMatrixBase< BaseFloat > &	out_value,
		const CuMatrixBase< BaseFloat > &	out_deriv,
		void *	memo,
		Component *	to_update,
		CuMatrixBase< BaseFloat > *	in_deriv
	)		const

virtual

Backprop function; depending on which of the arguments 'to_update' and 'in_deriv' are non-NULL, this can compute input-data derivatives and/or perform model update.

Parameters

[in]	debug_info	The component name, to be printed out in any warning messages.
[in]	indexes	A pointer to some information output by this class's PrecomputeIndexes function (will be NULL for simple components, i.e. those that don't do things like splicing).
[in]	in_value	The matrix that was given as input to the Propagate function. Will be ignored (and may be empty) if Properties()&kBackpropNeedsInput == 0.
[in]	out_value	The matrix that was output from the Propagate function. Will be ignored (and may be empty) if Properties()&kBackpropNeedsOutput == 0
[in]	out_deriv	The derivative at the output of this component.
[in]	memo	This will normally be NULL, but for component types that set the flag kUsesMemo, this will be the return value of the Propagate() function that corresponds to this Backprop() function. Ownership of any pointers is not transferred to the Backprop function; DeleteMemo() will be called to delete it.
[out]	to_update	If model update is desired, the Component to be updated, else NULL. Does not have to be identical to this. If supplied, you can assume that to_update->Properties() & kUpdatableComponent is nonzero.
[out]	in_deriv	The derivative at the input of this component, if needed (else NULL). If Properties()&kBackpropInPlace, may be the same matrix as out_deriv. If Properties()&kBackpropAdds, this is added to by the Backprop routine, else it is set. The component code chooses which mode to work in, based on convenience.

Implements Component.

Definition at line 443 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Properties(), LstmNonlinearityComponent::Properties(), and MaxpoolingComponent::Properties().

                                                                              {
   NVTX_RANGE("ConvolutionComponent::Backprop");
   ConvolutionComponent *to_update =
       dynamic_cast<ConvolutionComponent*>(to_update_in);
   const int32 num_x_steps = (1 + (input_x_dim_ - filt_x_dim_) / filt_x_step_),
               num_y_steps = (1 + (input_y_dim_ - filt_y_dim_) / filt_y_step_),
               num_filters = filter_params_.NumRows(),
               num_frames = out_deriv.NumRows(),
               filter_dim = filter_params_.NumCols();
 
   KALDI_ASSERT(out_deriv.NumRows() == num_frames &&
                out_deriv.NumCols() ==
                (num_filters * num_x_steps * num_y_steps));
 
   // Compute inderiv patches
   CuMatrix<BaseFloat> in_deriv_patches(num_frames,
                                        num_x_steps * num_y_steps * filter_dim,
                                        kSetZero);
 
   std::vector<CuSubMatrix<BaseFloat>* > patch_deriv_batch, out_deriv_batch,
       filter_params_batch;
   CuSubMatrix<BaseFloat>* filter_params_elem = new CuSubMatrix<BaseFloat>(
       filter_params_, 0, filter_params_.NumRows(), 0, filter_params_.NumCols());
 
   for (int32 x_step = 0; x_step < num_x_steps; x_step++)  {
     for (int32 y_step = 0; y_step < num_y_steps; y_step++)  {
       int32 patch_number = x_step * num_y_steps + y_step;
 
       patch_deriv_batch.push_back(new CuSubMatrix<BaseFloat>(
               in_deriv_patches.ColRange(
               patch_number * filter_dim, filter_dim)));
       out_deriv_batch.push_back(new CuSubMatrix<BaseFloat>(out_deriv.ColRange(
               patch_number * num_filters, num_filters)));
       filter_params_batch.push_back(filter_params_elem);
     }
   }
   AddMatMatBatched<BaseFloat>(1.0, patch_deriv_batch,
                               out_deriv_batch, kNoTrans,
                               filter_params_batch, kNoTrans, 0.0);
 
   if (in_deriv) {
     // combine the derivatives from the individual input deriv patches
     // to compute input deriv matrix
     InderivPatchesToInderiv(in_deriv_patches, in_deriv);
   }
 
   if (to_update != NULL)  {
     to_update->Update(debug_info, in_value, out_deriv, out_deriv_batch);
   }
 
   // release memory
   delete filter_params_elem;
   for (int32 p = 0; p < patch_deriv_batch.size(); p++) {
     delete patch_deriv_batch[p];
     delete out_deriv_batch[p];
   }
 }

◆ BiasParams()

const CuVector<BaseFloat>& BiasParams ( ) const

inline

Definition at line 179 of file nnet-combined-component.h.

References ConvolutionComponent::bias_params_.

179 { return bias_params_; }

kaldi::nnet3::ConvolutionComponent::bias_params_

CuVector< BaseFloat > bias_params_

Definition: nnet-combined-component.h:242

◆ Copy()

Component * Copy ( ) const

virtual

Copies component (deep copy).

Implements Component.

Definition at line 654 of file nnet-combined-component.cc.

References ConvolutionComponent::ConvolutionComponent().

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                             {
   ConvolutionComponent *ans = new ConvolutionComponent(*this);
   return ans;
 }

◆ DotProduct()

BaseFloat DotProduct ( const UpdatableComponent & other ) const

virtual

Computes dot-product between parameters of two instances of a Component.

Can be used for computing parameter-norm of an UpdatableComponent.

Implements UpdatableComponent.

Definition at line 647 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, ConvolutionComponent::filter_params_, kaldi::kTrans, kaldi::TraceMatMat(), and kaldi::VecVec().

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                                                                    {
   const ConvolutionComponent *other =
       dynamic_cast<const ConvolutionComponent*>(&other_in);
   return TraceMatMat(filter_params_, other->filter_params_, kTrans)
          + VecVec(bias_params_, other->bias_params_);
 }

◆ InderivPatchesToInderiv()

void InderivPatchesToInderiv	(	const CuMatrix< BaseFloat > &	in_deriv_patches,
		CuMatrixBase< BaseFloat > *	in_deriv
	)		const

private

Definition at line 387 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Backprop(), and MaxpoolingComponent::Copy().

                                              {
 
   const int32 num_x_steps = (1 + (input_x_dim_ - filt_x_dim_) / filt_x_step_),
               num_y_steps = (1 + (input_y_dim_ - filt_y_dim_) / filt_y_step_),
               filt_x_step = filt_x_step_,
               filt_y_step = filt_y_step_,
               filt_x_dim = filt_x_dim_,
               filt_y_dim = filt_y_dim_,
               input_x_dim = input_x_dim_,
               input_y_dim = input_y_dim_,
               input_z_dim = input_z_dim_,
               filter_dim = filter_params_.NumCols();
 
   // Compute the reverse column_map from the matrix with input
   // derivative patches to input derivative matrix
   std::vector<std::vector<int32> > reverse_column_map(in_deriv->NumCols());
   int32 rev_col_map_size = reverse_column_map.size();
   for (int32 x_step = 0; x_step < num_x_steps; x_step++) {
     for (int32 y_step = 0; y_step < num_y_steps; y_step++)  {
       int32 patch_number = x_step * num_y_steps + y_step;
       int32 patch_start_index = patch_number * filter_dim;
       for (int32 x = 0, index = patch_start_index; x < filt_x_dim; x++)  {
         for (int32 y = 0; y < filt_y_dim; y++)  {
           for (int32 z = 0; z < input_z_dim; z++, index++)  {
             int32 vector_index;
             if (input_vectorization_ == kZyx)  {
               vector_index = ZyxVectorIndex(x_step * filt_x_step + x,
                                             y_step * filt_y_step + y, z,
                                             input_x_dim, input_y_dim,
                                             input_z_dim);
             } else {
               KALDI_ASSERT(input_vectorization_ == kYzx);
               vector_index = YzxVectorIndex(x_step * filt_x_step + x,
                                             y_step * filt_y_step + y, z,
                                             input_x_dim, input_y_dim,
                                             input_z_dim);
             }
             KALDI_ASSERT(vector_index < rev_col_map_size);
             reverse_column_map[vector_index].push_back(index);
           }
         }
       }
     }
   }
   std::vector<std::vector<int32> > rearranged_column_map;
   RearrangeIndexes(reverse_column_map, &rearranged_column_map);
   for (int32 p = 0; p < rearranged_column_map.size(); p++) {
     CuArray<int32> cu_cols(rearranged_column_map[p]);
     in_deriv->AddCols(in_deriv_patches, cu_cols);
   }
 }

◆ Info()

std::string Info ( ) const

virtual

Returns some text-form information about this component, for diagnostics.

Starts with the type of the component. E.g. "SigmoidComponent dim=900", although most components will have much more info.

Reimplemented from UpdatableComponent.

Definition at line 147 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, ConvolutionComponent::filt_x_dim_, ConvolutionComponent::filt_x_step_, ConvolutionComponent::filt_y_dim_, ConvolutionComponent::filt_y_step_, ConvolutionComponent::filter_params_, UpdatableComponent::Info(), ConvolutionComponent::input_vectorization_, ConvolutionComponent::input_x_dim_, ConvolutionComponent::input_y_dim_, ConvolutionComponent::input_z_dim_, and kaldi::nnet3::PrintParameterStats().

Referenced by MaxpoolingComponent::MaxpoolingComponent().

                                            {
   std::ostringstream stream;
   stream << UpdatableComponent::Info()
          << ", input-x-dim=" << input_x_dim_
          << ", input-y-dim=" << input_y_dim_
          << ", input-z-dim=" << input_z_dim_
          << ", filt-x-dim=" << filt_x_dim_
          << ", filt-y-dim=" << filt_y_dim_
          << ", filt-x-step=" << filt_x_step_
          << ", filt-y-step=" << filt_y_step_
          << ", input-vectorization=" << input_vectorization_
          << ", num-filters=" << filter_params_.NumRows();
   PrintParameterStats(stream, "filter-params", filter_params_);
   PrintParameterStats(stream, "bias-params", bias_params_, true);
   return stream.str();
 }

◆ Init() [1/2]

void Init	(	int32	input_x_dim,
		int32	input_y_dim,
		int32	input_z_dim,
		int32	filt_x_dim,
		int32	filt_y_dim,
		int32	filt_x_step,
		int32	filt_y_step,
		int32	num_filters,
		TensorVectorizationType	input_vectorization,
		BaseFloat	param_stddev,
		BaseFloat	bias_stddev
	)

Definition at line 94 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, ConvolutionComponent::filt_x_dim_, ConvolutionComponent::filt_x_step_, ConvolutionComponent::filt_y_dim_, ConvolutionComponent::filt_y_step_, ConvolutionComponent::filter_params_, ConvolutionComponent::input_vectorization_, ConvolutionComponent::input_x_dim_, ConvolutionComponent::input_y_dim_, ConvolutionComponent::input_z_dim_, and KALDI_ASSERT.

Referenced by ConvolutionComponent::InitFromConfig(), ConvolutionComponent::LinearParams(), and LstmNonlinearityComponent::Properties().

                                                    {
   input_x_dim_ = input_x_dim;
   input_y_dim_ = input_y_dim;
   input_z_dim_ = input_z_dim;
   filt_x_dim_ = filt_x_dim;
   filt_y_dim_ = filt_y_dim;
   filt_x_step_ = filt_x_step;
   filt_y_step_ = filt_y_step;
   input_vectorization_ = input_vectorization;
   KALDI_ASSERT((input_x_dim_ - filt_x_dim_) % filt_x_step_ == 0);
   KALDI_ASSERT((input_y_dim_ - filt_y_dim_) % filt_y_step_ == 0);
   int32 filter_dim = filt_x_dim_ * filt_y_dim_ * input_z_dim_;
   filter_params_.Resize(num_filters, filter_dim);
   bias_params_.Resize(num_filters);
   KALDI_ASSERT(param_stddev >= 0.0 && bias_stddev >= 0.0);
   filter_params_.SetRandn();
   filter_params_.Scale(param_stddev);
   bias_params_.SetRandn();
   bias_params_.Scale(bias_stddev);
 }

◆ Init() [2/2]

void Init	(	int32	input_x_dim,
		int32	input_y_dim,
		int32	input_z_dim,
		int32	filt_x_dim,
		int32	filt_y_dim,
		int32	filt_x_step,
		int32	filt_y_step,
		TensorVectorizationType	input_vectorization,
		std::string	matrix_filename
	)

Definition at line 121 of file nnet-combined-component.cc.

                                {
   input_x_dim_ = input_x_dim;
   input_y_dim_ = input_y_dim;
   input_z_dim_ = input_z_dim;
   filt_x_dim_ = filt_x_dim;
   filt_y_dim_ = filt_y_dim;
   filt_x_step_ = filt_x_step;
   filt_y_step_ = filt_y_step;
   input_vectorization_ = input_vectorization;
   CuMatrix<BaseFloat> mat;
   ReadKaldiObject(matrix_filename, &mat);
   int32 filter_dim = (filt_x_dim_ * filt_y_dim_ * input_z_dim_);
   int32 num_filters = mat.NumRows();
   KALDI_ASSERT(mat.NumCols() == (filter_dim + 1));
   filter_params_.Resize(num_filters, filter_dim);
   bias_params_.Resize(num_filters);
   filter_params_.CopyFromMat(mat.Range(0, num_filters, 0, filter_dim));
   bias_params_.CopyColFromMat(mat, filter_dim);
 }

◆ InitFromConfig()

void InitFromConfig ( ConfigLine * cfl )

virtual

Initialize, from a ConfigLine object.

Parameters

[in] cfl A ConfigLine containing any parameters that are needed for initialization. For example: "dim=100 param-stddev=0.1"

Implements Component.

Definition at line 165 of file nnet-combined-component.cc.

References ConfigLine::GetValue(), ConfigLine::HasUnusedValues(), ConvolutionComponent::Init(), UpdatableComponent::InitLearningRatesFromConfig(), KALDI_ERR, ConvolutionComponent::kYzx, ConvolutionComponent::kZyx, ConfigLine::UnusedValues(), and ConfigLine::WholeLine().

Referenced by MaxpoolingComponent::MaxpoolingComponent().

                                                          {
   bool ok = true;
   std::string matrix_filename;
   int32 input_x_dim = -1, input_y_dim = -1, input_z_dim = -1,
         filt_x_dim = -1, filt_y_dim = -1,
         filt_x_step = -1, filt_y_step = -1,
         num_filters = -1;
   std::string input_vectorization_order = "zyx";
   InitLearningRatesFromConfig(cfl);
   ok = ok && cfl->GetValue("input-x-dim", &input_x_dim);
   ok = ok && cfl->GetValue("input-y-dim", &input_y_dim);
   ok = ok && cfl->GetValue("input-z-dim", &input_z_dim);
   ok = ok && cfl->GetValue("filt-x-dim", &filt_x_dim);
   ok = ok && cfl->GetValue("filt-y-dim", &filt_y_dim);
   ok = ok && cfl->GetValue("filt-x-step", &filt_x_step);
   ok = ok && cfl->GetValue("filt-y-step", &filt_y_step);
 
   if (!ok)
     KALDI_ERR << "Bad initializer " << cfl->WholeLine();
   // optional argument
   TensorVectorizationType input_vectorization;
   cfl->GetValue("input-vectorization-order", &input_vectorization_order);
   if (input_vectorization_order.compare("zyx") == 0) {
     input_vectorization = kZyx;
   } else if (input_vectorization_order.compare("yzx") == 0) {
     input_vectorization = kYzx;
   } else {
     KALDI_ERR << "Unknown or unsupported input vectorization order "
               << input_vectorization_order
               << " accepted candidates are 'yzx' and 'zyx'";
   }
 
   if (cfl->GetValue("matrix", &matrix_filename)) {
     // initialize from prefined parameter matrix
     Init(input_x_dim, input_y_dim, input_z_dim,
          filt_x_dim, filt_y_dim,
          filt_x_step, filt_y_step,
          input_vectorization,
          matrix_filename);
   } else {
     ok = ok && cfl->GetValue("num-filters", &num_filters);
     if (!ok)
       KALDI_ERR << "Bad initializer " << cfl->WholeLine();
     // initialize from configuration
     int32 filter_input_dim = filt_x_dim * filt_y_dim * input_z_dim;
     BaseFloat param_stddev = 1.0 / std::sqrt(filter_input_dim), bias_stddev = 1.0;
     cfl->GetValue("param-stddev", &param_stddev);
     cfl->GetValue("bias-stddev", &bias_stddev);
     Init(input_x_dim, input_y_dim, input_z_dim,
          filt_x_dim, filt_y_dim, filt_x_step, filt_y_step, num_filters,
          input_vectorization, param_stddev, bias_stddev);
   }
   if (cfl->HasUnusedValues())
     KALDI_ERR << "Could not process these elements in initializer: "
               << cfl->UnusedValues();
   if (!ok)
     KALDI_ERR << "Bad initializer " << cfl->WholeLine();
 }

◆ InputDim()

int32 InputDim ( ) const

virtual

Returns input-dimension of this component.

Implements Component.

Definition at line 81 of file nnet-combined-component.cc.

References ConvolutionComponent::input_x_dim_, ConvolutionComponent::input_y_dim_, and ConvolutionComponent::input_z_dim_.

Referenced by MaxpoolingComponent::MaxpoolingComponent().

                                            {
   return input_x_dim_ * input_y_dim_ * input_z_dim_;
 }

◆ InputToInputPatches()

void InputToInputPatches	(	const CuMatrixBase< BaseFloat > &	in,
		CuMatrix< BaseFloat > *	patches
	)		const

private

Definition at line 245 of file nnet-combined-component.cc.

Referenced by MaxpoolingComponent::Copy(), ConvolutionComponent::Propagate(), and ConvolutionComponent::Update().

                                        {
   int32 num_x_steps = (1 + (input_x_dim_ - filt_x_dim_) / filt_x_step_);
   int32 num_y_steps = (1 + (input_y_dim_ - filt_y_dim_) / filt_y_step_);
   const int32 filt_x_step = filt_x_step_,
               filt_y_step = filt_y_step_,
               filt_x_dim = filt_x_dim_,
               filt_y_dim = filt_y_dim_,
               input_x_dim = input_x_dim_,
               input_y_dim = input_y_dim_,
               input_z_dim = input_z_dim_,
               filter_dim = filter_params_.NumCols();
 
   std::vector<int32> column_map(patches->NumCols());
   int32 column_map_size = column_map.size();
   for (int32 x_step = 0; x_step < num_x_steps; x_step++) {
     for (int32 y_step = 0; y_step < num_y_steps; y_step++)  {
       int32 patch_number = x_step * num_y_steps + y_step;
       int32 patch_start_index = patch_number * filter_dim;
       for (int32 x = 0, index = patch_start_index; x < filt_x_dim; x++)  {
         for (int32 y = 0; y < filt_y_dim; y++)  {
           for (int32 z = 0; z < input_z_dim; z++, index++)  {
             KALDI_ASSERT(index < column_map_size);
             if (input_vectorization_ == kZyx)  {
               column_map[index] = ZyxVectorIndex(x_step * filt_x_step + x,
                                                  y_step * filt_y_step + y, z,
                                                  input_x_dim, input_y_dim,
                                                  input_z_dim);
             } else if (input_vectorization_ == kYzx)  {
               column_map[index] = YzxVectorIndex(x_step * filt_x_step + x,
                                                   y_step * filt_y_step + y, z,
                                                   input_x_dim, input_y_dim,
                                                   input_z_dim);
             }
           }
         }
       }
     }
   }
   CuArray<int32> cu_cols(column_map);
   patches->CopyCols(in, cu_cols);
 }

◆ LinearParams()

const CuMatrix<BaseFloat>& LinearParams ( ) const

inline

Definition at line 180 of file nnet-combined-component.h.

References ConvolutionComponent::filter_params_, ConvolutionComponent::Init(), ConvolutionComponent::Resize(), and ConvolutionComponent::Update().

180 { return filter_params_; }

kaldi::nnet3::ConvolutionComponent::filter_params_

CuMatrix< BaseFloat > filter_params_

Definition: nnet-combined-component.h:234

◆ NumParameters()

int32 NumParameters ( ) const

virtual

The following new virtual function returns the total dimension of the parameters in this class.

Reimplemented from UpdatableComponent.

Definition at line 676 of file nnet-combined-component.cc.

References ConvolutionComponent::filter_params_.

Referenced by ConvolutionComponent::Properties(), LstmNonlinearityComponent::Properties(), ConvolutionComponent::UnVectorize(), and ConvolutionComponent::Vectorize().

                                                 {
   return (filter_params_.NumCols() + 1) * filter_params_.NumRows();
 }

◆ operator=()

const ConvolutionComponent& operator= ( const ConvolutionComponent & other )

private

◆ OutputDim()

int32 OutputDim ( ) const

virtual

Returns output-dimension of this component.

Implements Component.

Definition at line 86 of file nnet-combined-component.cc.

References ConvolutionComponent::filt_x_dim_, ConvolutionComponent::filt_x_step_, ConvolutionComponent::filt_y_dim_, ConvolutionComponent::filt_y_step_, ConvolutionComponent::filter_params_, ConvolutionComponent::input_x_dim_, and ConvolutionComponent::input_y_dim_.

Referenced by MaxpoolingComponent::MaxpoolingComponent().

                                             {
   int32 num_x_steps = (1 + (input_x_dim_ - filt_x_dim_) / filt_x_step_);
   int32 num_y_steps = (1 + (input_y_dim_ - filt_y_dim_) / filt_y_step_);
   int32 num_filters = filter_params_.NumRows();
   return num_x_steps * num_y_steps * num_filters;
 }

◆ PerturbParams()

void PerturbParams ( BaseFloat stddev )

virtual

This function is to be used in testing.

It adds unit noise times "stddev" to the parameters of the component.

Implements UpdatableComponent.

Definition at line 659 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, ConvolutionComponent::filter_params_, CuVectorBase< Real >::SetRandn(), and CuMatrixBase< Real >::SetRandn().

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                                          {
   CuMatrix<BaseFloat> temp_filter_params(filter_params_);
   temp_filter_params.SetRandn();
   filter_params_.AddMat(stddev, temp_filter_params);
 
   CuVector<BaseFloat> temp_bias_params(bias_params_);
   temp_bias_params.SetRandn();
   bias_params_.AddVec(stddev, temp_bias_params);
 }

◆ Propagate()

void * Propagate	(	const ComponentPrecomputedIndexes *	indexes,
		const CuMatrixBase< BaseFloat > &	in,
		CuMatrixBase< BaseFloat > *	out
	)		const

virtual

Propagate function.

Parameters

[in]	indexes	A pointer to some information output by this class's PrecomputeIndexes function (will be NULL for simple components, i.e. those that don't do things like splicing).
[in]	in	The input to this component. Num-columns == InputDim().
[out]	out	The output of this component. Num-columns == OutputDim(). Note: output of this component will be added to the initial value of "out" if Properties()&kPropagateAdds != 0; otherwise the output will be set and the initial value ignored. Each Component chooses whether it is more convenient implementation-wise to add or set, and the calling code has to deal with it.

Returns: Normally returns NULL, but may return a non-NULL value for components which have the flag kUsesMemo set. This value will be passed into the corresponding Backprop routine.

Implements Component.

Definition at line 292 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Properties(), LstmNonlinearityComponent::Properties(), and MaxpoolingComponent::Properties().

                                                                              {
   const int32 num_x_steps = (1 + (input_x_dim_ - filt_x_dim_) / filt_x_step_),
               num_y_steps = (1 + (input_y_dim_ - filt_y_dim_) / filt_y_step_),
               num_filters = filter_params_.NumRows(),
               num_frames = in.NumRows(),
               filter_dim = filter_params_.NumCols();
   KALDI_ASSERT((*out).NumRows() == num_frames &&
                (*out).NumCols() == (num_filters * num_x_steps * num_y_steps));
 
   CuMatrix<BaseFloat> patches(num_frames,
                               num_x_steps * num_y_steps * filter_dim,
                               kUndefined);
   InputToInputPatches(in, &patches);
   CuSubMatrix<BaseFloat>* filter_params_elem = new CuSubMatrix<BaseFloat>(
       filter_params_, 0, filter_params_.NumRows(), 0, filter_params_.NumCols());
   std::vector<CuSubMatrix<BaseFloat>* > tgt_batch, patch_batch,
       filter_params_batch;
 
   for (int32 x_step = 0; x_step < num_x_steps; x_step++)  {
     for (int32 y_step = 0; y_step < num_y_steps; y_step++)  {
       int32 patch_number = x_step * num_y_steps + y_step;
       tgt_batch.push_back(new CuSubMatrix<BaseFloat>(
               out->ColRange(patch_number * num_filters, num_filters)));
       patch_batch.push_back(new CuSubMatrix<BaseFloat>(
               patches.ColRange(patch_number * filter_dim, filter_dim)));
       filter_params_batch.push_back(filter_params_elem);
       tgt_batch[patch_number]->AddVecToRows(1.0, bias_params_, 1.0); // add bias
     }
   }
   // apply all filters
   AddMatMatBatched<BaseFloat>(1.0, tgt_batch, patch_batch,
                               kNoTrans, filter_params_batch,
                               kTrans, 1.0);
   // release memory
   delete filter_params_elem;
   for (int32 p = 0; p < tgt_batch.size(); p++) {
     delete tgt_batch[p];
     delete patch_batch[p];
   }
   return NULL;
 }

◆ Properties()

virtual int32 Properties ( ) const

inlinevirtual

Return bitmask of the component's properties.

These properties depend only on the component's type. See enum ComponentProperties.

Implements Component.

Definition at line 140 of file nnet-combined-component.h.

                                    {
     return kSimpleComponent|kUpdatableComponent|kBackpropNeedsInput|
            kBackpropAdds|kPropagateAdds;
   }

◆ Read()

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

virtual

Read function (used after we know the type of the Component); accepts input that is missing the token that describes the component type, in case it has already been consumed.

Implements Component.

Definition at line 585 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Properties(), LstmNonlinearityComponent::Properties(), and MaxpoolingComponent::Properties().

                                                            {
   ReadUpdatableCommon(is, binary);  // Read opening tag and learning rate.
   ExpectToken(is, binary, "<InputXDim>");
   ReadBasicType(is, binary, &input_x_dim_);
   ExpectToken(is, binary, "<InputYDim>");
   ReadBasicType(is, binary, &input_y_dim_);
   ExpectToken(is, binary, "<InputZDim>");
   ReadBasicType(is, binary, &input_z_dim_);
   ExpectToken(is, binary, "<FiltXDim>");
   ReadBasicType(is, binary, &filt_x_dim_);
   ExpectToken(is, binary, "<FiltYDim>");
   ReadBasicType(is, binary, &filt_y_dim_);
   ExpectToken(is, binary, "<FiltXStep>");
   ReadBasicType(is, binary, &filt_x_step_);
   ExpectToken(is, binary, "<FiltYStep>");
   ReadBasicType(is, binary, &filt_y_step_);
   ExpectToken(is, binary, "<InputVectorization>");
   int32 input_vectorization;
   ReadBasicType(is, binary, &input_vectorization);
   input_vectorization_ = static_cast<TensorVectorizationType>(input_vectorization);
   ExpectToken(is, binary, "<FilterParams>");
   filter_params_.Read(is, binary);
   ExpectToken(is, binary, "<BiasParams>");
   bias_params_.Read(is, binary);
   std::string tok;
   ReadToken(is, binary, &tok);
   if (tok == "<IsGradient>") {
     ReadBasicType(is, binary, &is_gradient_);
     ExpectToken(is, binary, "</ConvolutionComponent>");
   } else {
     is_gradient_ = false;
     KALDI_ASSERT(tok == "</ConvolutionComponent>");
   }
 }

◆ Resize()

void Resize	(	int32	input_dim,
		int32	output_dim
	)

Referenced by ConvolutionComponent::LinearParams().

◆ Scale()

void Scale ( BaseFloat scale )

virtual

This virtual function when called on – an UpdatableComponent scales the parameters by "scale" when called by an UpdatableComponent.

– a Nonlinear component (or another component that stores stats, like BatchNormComponent)– it relates to scaling activation stats, not parameters. Otherwise it will normally do nothing.

Reimplemented from Component.

Definition at line 337 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, and ConvolutionComponent::filter_params_.

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                                 {
   if (scale == 0.0) {
     filter_params_.SetZero();
     bias_params_.SetZero();
   } else {
     filter_params_.Scale(scale);
     bias_params_.Scale(scale);
   }
 }

◆ SetParams()

void SetParams	(	const VectorBase< BaseFloat > &	bias,
		const MatrixBase< BaseFloat > &	filter
	)

Definition at line 669 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, ConvolutionComponent::filter_params_, and KALDI_ASSERT.

Referenced by ConvolutionComponent::Properties().

                                                                           {
   bias_params_ = bias;
   filter_params_ = filter;
   KALDI_ASSERT(bias_params_.Dim() == filter_params_.NumRows());
 }

◆ Type()

virtual std::string Type ( ) const

inlinevirtual

Returns a string such as "SigmoidComponent", describing the type of the object.

Implements Component.

Definition at line 139 of file nnet-combined-component.h.

139 { return "ConvolutionComponent"; }

◆ UnVectorize()

void UnVectorize ( const VectorBase< BaseFloat > & params )

virtual

Converts the parameters from vector form.

Reimplemented from UpdatableComponent.

Definition at line 686 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, VectorBase< Real >::Dim(), ConvolutionComponent::filter_params_, KALDI_ASSERT, ConvolutionComponent::NumParameters(), and VectorBase< Real >::Range().

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                                                           {
   KALDI_ASSERT(params.Dim() == this->NumParameters());
   int32 num_filter_params = filter_params_.NumCols() * filter_params_.NumRows();
   filter_params_.CopyRowsFromVec(params.Range(0, num_filter_params));
   bias_params_.CopyFromVec(params.Range(num_filter_params, bias_params_.Dim()));
 }

◆ Update() [1/2]

void Update	(	const std::string &	debug_info,
		const CuMatrixBase< BaseFloat > &	in_value,
		const CuMatrixBase< BaseFloat > &	out_deriv,
		const std::vector< CuSubMatrix< BaseFloat > *> &	out_deriv_batch
	)

Definition at line 511 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::LinearParams(), and ConvolutionComponent::Properties().

                                                                                               {
   // useful dims
   const int32 num_x_steps = (1 + (input_x_dim_ - filt_x_dim_) / filt_x_step_),
               num_y_steps = (1 + (input_y_dim_ - filt_y_dim_) / filt_y_step_),
               num_filters = filter_params_.NumRows(),
               num_frames = out_deriv.NumRows(),
               filter_dim = filter_params_.NumCols();
   KALDI_ASSERT(out_deriv.NumRows() == num_frames &&
                out_deriv.NumCols() ==
                (num_filters * num_x_steps * num_y_steps));
 
 
   CuMatrix<BaseFloat> filters_grad;
   CuVector<BaseFloat> bias_grad;
 
   CuMatrix<BaseFloat> input_patches(num_frames,
                                     filter_dim * num_x_steps * num_y_steps,
                                     kUndefined);
   InputToInputPatches(in_value, &input_patches);
 
   filters_grad.Resize(num_filters, filter_dim, kSetZero); // reset
   bias_grad.Resize(num_filters, kSetZero); // reset
 
   // create a single large matrix holding the smaller matrices
   // from the vector container filters_grad_batch along the rows
   CuMatrix<BaseFloat> filters_grad_blocks_batch(
       num_x_steps * num_y_steps * filters_grad.NumRows(),
       filters_grad.NumCols());
 
   std::vector<CuSubMatrix<BaseFloat>* > filters_grad_batch, input_patch_batch;
 
   for (int32 x_step = 0; x_step < num_x_steps; x_step++)  {
     for (int32 y_step = 0; y_step < num_y_steps; y_step++)  {
       int32 patch_number = x_step * num_y_steps + y_step;
       filters_grad_batch.push_back(new CuSubMatrix<BaseFloat>(
           filters_grad_blocks_batch.RowRange(
               patch_number * filters_grad.NumRows(), filters_grad.NumRows())));
 
       input_patch_batch.push_back(new CuSubMatrix<BaseFloat>(
               input_patches.ColRange(patch_number * filter_dim, filter_dim)));
     }
   }
 
   AddMatMatBatched<BaseFloat>(1.0, filters_grad_batch, out_deriv_batch, kTrans,
                               input_patch_batch, kNoTrans, 1.0);
 
   // add the row blocks together to filters_grad
   filters_grad.AddMatBlocks(1.0, filters_grad_blocks_batch);
 
   // create a matrix holding the col blocks sum of out_deriv
   CuMatrix<BaseFloat> out_deriv_col_blocks_sum(out_deriv.NumRows(),
                                                num_filters);
 
   // add the col blocks together to out_deriv_col_blocks_sum
   out_deriv_col_blocks_sum.AddMatBlocks(1.0, out_deriv);
 
   bias_grad.AddRowSumMat(1.0, out_deriv_col_blocks_sum, 1.0);
 
   // release memory
   for (int32 p = 0; p < input_patch_batch.size(); p++) {
     delete filters_grad_batch[p];
     delete input_patch_batch[p];
   }
 
   //
   // update
   //
   filter_params_.AddMat(learning_rate_, filters_grad);
   bias_params_.AddVec(learning_rate_, bias_grad);
 }

◆ Update() [2/2]

void Update	(	const std::string &	debug_info,
		const CuMatrixBase< BaseFloat > &	in_value,
		const CuMatrixBase< BaseFloat > &	out_deriv
	)

◆ Vectorize()

void Vectorize ( VectorBase< BaseFloat > * params ) const

virtual

Turns the parameters into vector form.

We put the vector form on the CPU, because in the kinds of situations where we do this, we'll tend to use too much memory for the GPU.

Reimplemented from UpdatableComponent.

Definition at line 680 of file nnet-combined-component.cc.

References ConvolutionComponent::bias_params_, VectorBase< Real >::Dim(), ConvolutionComponent::filter_params_, KALDI_ASSERT, ConvolutionComponent::NumParameters(), and VectorBase< Real >::Range().

Referenced by ConvolutionComponent::Properties(), and LstmNonlinearityComponent::Properties().

                                                                         {
   KALDI_ASSERT(params->Dim() == this->NumParameters());
   int32 num_filter_params = filter_params_.NumCols() * filter_params_.NumRows();
   params->Range(0, num_filter_params).CopyRowsFromMat(filter_params_);
   params->Range(num_filter_params, bias_params_.Dim()).CopyFromVec(bias_params_);
 }

◆ Write()

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

virtual

Write component to stream.

Implements Component.

Definition at line 620 of file nnet-combined-component.cc.

Referenced by ConvolutionComponent::Properties(), LstmNonlinearityComponent::Properties(), and MaxpoolingComponent::Properties().

                                                                   {
   WriteUpdatableCommon(os, binary);  // write opening tag and learning rate.
   WriteToken(os, binary, "<InputXDim>");
   WriteBasicType(os, binary, input_x_dim_);
   WriteToken(os, binary, "<InputYDim>");
   WriteBasicType(os, binary, input_y_dim_);
   WriteToken(os, binary, "<InputZDim>");
   WriteBasicType(os, binary, input_z_dim_);
   WriteToken(os, binary, "<FiltXDim>");
   WriteBasicType(os, binary, filt_x_dim_);
   WriteToken(os, binary, "<FiltYDim>");
   WriteBasicType(os, binary, filt_y_dim_);
   WriteToken(os, binary, "<FiltXStep>");
   WriteBasicType(os, binary, filt_x_step_);
   WriteToken(os, binary, "<FiltYStep>");
   WriteBasicType(os, binary, filt_y_step_);
   WriteToken(os, binary, "<InputVectorization>");
   WriteBasicType(os, binary, static_cast<int32>(input_vectorization_));
   WriteToken(os, binary, "<FilterParams>");
   filter_params_.Write(os, binary);
   WriteToken(os, binary, "<BiasParams>");
   bias_params_.Write(os, binary);
   WriteToken(os, binary, "<IsGradient>");
   WriteBasicType(os, binary, is_gradient_);
   WriteToken(os, binary, "</ConvolutionComponent>");
 }

Member Data Documentation

◆ bias_params_

CuVector<BaseFloat> bias_params_

private

Definition at line 242 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Add(), ConvolutionComponent::BiasParams(), ConvolutionComponent::DotProduct(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::PerturbParams(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Scale(), ConvolutionComponent::SetParams(), ConvolutionComponent::UnVectorize(), ConvolutionComponent::Update(), ConvolutionComponent::Vectorize(), and ConvolutionComponent::Write().

◆ filt_x_dim_

int32 filt_x_dim_

private

Definition at line 214 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::OutputDim(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Update(), and ConvolutionComponent::Write().

◆ filt_x_step_

int32 filt_x_step_

private

Definition at line 221 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::OutputDim(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Update(), and ConvolutionComponent::Write().

◆ filt_y_dim_

int32 filt_y_dim_

private

Definition at line 216 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::OutputDim(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Update(), and ConvolutionComponent::Write().

◆ filt_y_step_

int32 filt_y_step_

private

Definition at line 225 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::OutputDim(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Update(), and ConvolutionComponent::Write().

◆ filter_params_

CuMatrix<BaseFloat> filter_params_

private

◆ input_vectorization_

TensorVectorizationType input_vectorization_

private

Definition at line 231 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::Read(), and ConvolutionComponent::Write().

◆ input_x_dim_

int32 input_x_dim_

private

Definition at line 204 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputDim(), MaxpoolingComponent::InputDim(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::OutputDim(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Update(), and ConvolutionComponent::Write().

◆ input_y_dim_

int32 input_y_dim_

private

Definition at line 207 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::Backprop(), ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputDim(), MaxpoolingComponent::InputDim(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::OutputDim(), ConvolutionComponent::Propagate(), ConvolutionComponent::Read(), ConvolutionComponent::Update(), and ConvolutionComponent::Write().

◆ input_z_dim_

int32 input_z_dim_

private

Definition at line 210 of file nnet-combined-component.h.

Referenced by ConvolutionComponent::InderivPatchesToInderiv(), ConvolutionComponent::Info(), ConvolutionComponent::Init(), ConvolutionComponent::InputDim(), MaxpoolingComponent::InputDim(), ConvolutionComponent::InputToInputPatches(), ConvolutionComponent::Read(), and ConvolutionComponent::Write().

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

nnet3/nnet-combined-component.h
nnet3/nnet-combined-component.cc

Public Types

Public Member Functions

Private Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Member Enumeration Documentation

◆ TensorVectorizationType

Constructor & Destructor Documentation

◆ ConvolutionComponent() [1/3]

◆ ConvolutionComponent() [2/3]

◆ ConvolutionComponent() [3/3]

Member Function Documentation

◆ Add()

◆ Backprop()

◆ BiasParams()

◆ Copy()

◆ DotProduct()

◆ InderivPatchesToInderiv()

◆ Info()

◆ Init() [1/2]

◆ Init() [2/2]

◆ InitFromConfig()

◆ InputDim()

◆ InputToInputPatches()

◆ LinearParams()

◆ NumParameters()

◆ operator=()

◆ OutputDim()

◆ PerturbParams()

◆ Propagate()

◆ Properties()

◆ Read()

◆ Resize()

◆ Scale()

◆ SetParams()

◆ Type()

◆ UnVectorize()

◆ Update() [1/2]

◆ Update() [2/2]

◆ Vectorize()

◆ Write()

Member Data Documentation

◆ bias_params_

◆ filt_x_dim_

◆ filt_x_step_

◆ filt_y_dim_

◆ filt_y_step_

◆ filter_params_

◆ input_vectorization_

◆ input_x_dim_

◆ input_y_dim_

◆ input_z_dim_