Base layer class. More...

#include <layer.h>

Inheritance diagram for singa::Layer:

Public Member Functions
virtual void	Setup (const LayerProto &proto, int npartitions=1)
	Setup layer properties. More...

virtual void	ComputeFeature (int flag, Metric *perf)=0
	Compute features of this layer based on connected layers. More...

virtual void	ComputeGradient (int flag, Metric *perf)=0
	Compute gradients for parameters and connected layers. More...

virtual const std::vector < Param * >	GetParams () const
	Layers that have paramters must override this function. More...

virtual ConnectionType	src_neuron_connection (int k) const
	Return the connection type between one neuron of this layer and its source layer. More...

virtual ConnectionType	dst_layer_connection () const
	Return the connection type of this layer and all dst layers. More...

virtual const std::string	DebugString (int step, int flag)
	For print debug info about each layer, e.g., norm of feature vector, norm of parameters. More...

int	partition_dim () const

int	partition_id () const

int	type () const

const std::string &	name () const
	Return name of this layer.

virtual const Blob< float > &	data (const Layer *from) const

virtual Blob< float > *	mutable_data (const Layer *from)

virtual const Blob< float > &	grad (const Layer *from) const

virtual Blob< float > *	mutable_grad (const Layer *from)

const std::vector< Layer * >	srclayers () const
	return LayerS that connected to this layer

const std::vector< Layer * >	dstlayers () const
	return LayerS that this layer connected to

int	srclayers_size () const

int	dstlayers_size () const

void	clear_dstlayers ()

void	clear_srclayers ()

void	add_srclayer (Layer *src)

void	add_dstlayer (Layer *dst)

Static Public Member Functions
static Layer *	Create (const LayerProto &proto)

Protected Attributes
LayerProto	layer_proto_

Blob< float >	data_

Blob< float >	grad_

std::vector< Layer * >	srclayers_

std::vector< Layer * >	dstlayers_

Detailed Description

Base layer class.

Children should implement at least Layer::ComputeFeature() and Layer::ComputGradient() functions for contrastive-divergence/back-propagation algorithm.

Member Function Documentation

virtual void singa::Layer::ComputeFeature	(	int	flag,
		Metric *	perf
	)

pure virtual

Compute features of this layer based on connected layers.

Parameters

perf	pointer to Metric obj for collect and aggregate performance

Implemented in singa::RBMHidLayer, singa::RBMVisLayer, singa::PrefetchLayer, singa::SigmoidLayer, singa::STanhLayer, singa::InnerProductLayer, singa::SplitLayer, singa::ReLULayer, singa::CPoolingLayer, singa::SliceLayer, singa::ParserLayer, singa::PoolingLayer, singa::ConcateLayer, singa::ShardDataLayer, singa::BridgeSrcLayer, singa::DropoutLayer, singa::CConvolutionLayer, singa::BridgeDstLayer, singa::SoftmaxLossLayer, singa::ConvolutionLayer, and singa::EuclideanLossLayer.

virtual void singa::Layer::ComputeGradient	(	int	flag,
		Metric *	perf
	)

pure virtual

Compute gradients for parameters and connected layers.

Parameters

flag	used to get the calling phase, e.g., forward of training (kForward \| kTrain)
flag	used to get the calling phase, e.g., forward of training

Implemented in singa::RBMHidLayer, singa::RBMVisLayer, singa::PrefetchLayer, singa::SigmoidLayer, singa::STanhLayer, singa::InnerProductLayer, singa::SplitLayer, singa::ReLULayer, singa::CPoolingLayer, singa::SliceLayer, singa::ParserLayer, singa::PoolingLayer, singa::ConcateLayer, singa::BridgeSrcLayer, singa::DropoutLayer, singa::CConvolutionLayer, singa::BridgeDstLayer, singa::DataLayer, singa::SoftmaxLossLayer, singa::ConvolutionLayer, and singa::EuclideanLossLayer.

virtual const Blob<float>& singa::Layer::data ( const Layer * from ) const

inlinevirtual

Returns: name of src data blob, used by prefetch layer to locate the data blob in parser layers; The default value is "unknown"; If the src layer is the prefetch layer and there are more than one parser layers, this value be set. const std::string &datablob() const { return layer_proto_.datablob(); }; a const ref for Blob storing neuron values of this layer for BP

Reimplemented in singa::BridgeSrcLayer.

virtual const std::string singa::Layer::DebugString	(	int	step,
		int	flag
	)

virtual

For print debug info about each layer, e.g., norm of feature vector, norm of parameters.

Parameters

step	training/test/validation step
flag	used to get the calling phase, e.g., forward of training (kForward \| kTrain)

Returns: debug info about this layer.

virtual ConnectionType singa::Layer::dst_layer_connection ( ) const

inlinevirtual

Return the connection type of this layer and all dst layers.

Currently support two connection types: kOneToOne, and kOneToMany. kOneToOne indicates the users implement the ComputeFeature and ComputeGradient function considering only one dest layer. In this case, a SplitLayer will be added automatically to connect this layer with all dest layer. kOneToMany indicates the users has already considered multiple dest layers in the implementation.

Returns: connection type default is kOneToOne.

Reimplemented in singa::ParserLayer, and singa::DataLayer.

virtual const std::vector<Param*> singa::Layer::GetParams ( ) const

inlinevirtual

Layers that have paramters must override this function.

Parameters

flag	used to get the calling phase, e.g., forward of training (kForward \| kTrain)

Returns: parameters associated with this layer

Reimplemented in singa::RBMLayer, singa::InnerProductLayer, and singa::ConvolutionLayer.

virtual Blob<float>* singa::Layer::mutable_grad ( const Layer * from )

inlinevirtual

Returns: a pointer to storing neuron grads of this layer for BP

Reimplemented in singa::LossLayer, singa::ParserLayer, singa::BridgeSrcLayer, and singa::DataLayer.

int singa::Layer::partition_dim ( ) const

inline

Returns: partition dimension of this layer. -1 for no partition; 0 for partition the mini-batch into sub-mini-batch. 1 for partition the layer feature vector into sub-vector.

virtual void singa::Layer::Setup	(	const LayerProto &	proto,
		int	npartitions = `1`
	)

inlinevirtual

Setup layer properties.

Setup the shapes for data and parameters, also setup some properties based on the layer configuration and connected layers.

Parameters

proto	layer configuration.
npartitions	num of total partitions of the original layer. This layer should be setup as one partition.

Reimplemented in singa::RBMHidLayer, singa::RBMVisLayer, singa::RBMLayer, singa::SigmoidLayer, singa::RGBImageLayer, singa::STanhLayer, singa::MnistLayer, singa::LabelLayer, singa::InnerProductLayer, singa::SplitLayer, singa::ReLULayer, singa::CPoolingLayer, singa::SliceLayer, singa::PoolingLayer, singa::ConcateLayer, singa::ShardDataLayer, singa::DropoutLayer, singa::BridgeDstLayer, singa::SoftmaxLossLayer, and singa::ConvolutionLayer.

virtual ConnectionType singa::Layer::src_neuron_connection ( int k ) const

inlinevirtual

Return the connection type between one neuron of this layer and its source layer.

Currently support two connection types: kOneToOne, and kOneToAll. kOneToOne indicates the neuron depends on only one neuron from src layer. kOneToAll indicates the neuron depends on all neurons from src layer. TODO(wangwei) support kOneToMany.

Parameters

k	index of source layer (current only support k = 0.
connection	type.

Reimplemented in singa::SoftmaxLossLayer, and singa::ConvolutionLayer.

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

/home/wangwei/program/asf/release-0.1/apache-singa-incubating-0.1.0-RC1/include/neuralnet/layer.h

Public Member Functions

Static Public Member Functions

Protected Attributes

Detailed Description

Member Function Documentation