Inheritance diagram for MultiGPUConvNet:

Public Member Functions
	MultiGPUConvNet (const string &model_file)

Public Member Functions inherited from ConvNet
	ConvNet (const string &model_file)
	Instantiate a model using the config in model_file.

void	SetupDataset (const string &train_data_config_file)

virtual void	SetupDataset (const string &train_data_config_file, const string &val_data_config_file)

virtual void	Train ()
	Start training. More...

void	Validate (DataHandler *dataset, vector< float > &error)
	Validate the model on the specfied dataset and return the error. More...

void	Validate (vector< float > &error)
	Validate the model on the validation dataset and return the error. More...

void	Save ()
	Write the model to disk. More...

void	Save (const string &output_file)
	Write the model to disk in the file specified. More...

void	Load ()
	Load the model. More...

void	Load (const string &input_file)
	Load the model from the file specified. More...

void	Display ()
	Display the state of the model. More...

void	DumpOutputs (const string &output_file, DataHandler *dataset, const vector< string > &layer_names)
	Write the state of the layers to disk. More...

void	DumpOutputs (const string &output_file, const vector< string > &layer_names)
	Write the state of the layers to disk. More...

void	AllocateMemory (bool fprop_only)
	Allocate memory for the model. More...

Protected Member Functions
virtual void	Fprop (bool train)
	Forward propagate through the network. More...

virtual void	Bprop (bool update_weights)
	Backpropagate through the network and update weights. More...

Protected Member Functions inherited from ConvNet
void	BuildNet ()
	Creates layers and edges. More...

void	DestroyNet ()
	Release all memory held by the model. More...

void	AllocateLayerMemory ()
	Allocate layer memory for using mini-batches of batch_size_. More...

void	AllocateEdgeMemory (bool fprop_only)
	Allocate memory for edges. More...

string	GetCheckpointFilename ()

void	TimestampModel ()

void	Sort ()
	Topologically sort layers. More...

void	Fprop (Layer &input, Layer &output, Edge &edge, bool overwrite)
	Forward propagate one layer. More...

virtual void	Bprop (Layer &output, Layer &input, Edge &edge, bool overwrite, bool update_weights)
	Back propagate through one layer. More...

virtual void	ComputeDeriv ()
	Computes the derivative of the loss function. More...

virtual void	GetLoss (vector< float > &error)
	Computes the loss function (to be displayed). More...

void	DumpOutputs (const string &output_file, DataHandler dataset, vector< Layer > &layers)
	Write the state of the layers to disk. More...

virtual void	TrainOneBatch (vector< float > &error)
	Takes one optimization step. More...

void	DisplayLayers ()

void	DisplayEdges ()

void	InsertPolyak ()

void	LoadPolyakWeights ()

void	LoadCurrentWeights ()

void	WriteLog (int current_iter, float time, float training_error)

void	WriteLog (int current_iter, float time, const vector< float > &training_error)

void	WriteValLog (int current_iter, const vector< float > &error)

Layer *	GetLayerByName (const string &name)

bool	CheckReduceLearningRate (const vector< float > &val_error)
	Decides if learning rate should be reduced. More...

void	ReduceLearningRate (const float factor)
	Multiply learning rate by factor. More...

Additional Inherited Members
Static Protected Member Functions inherited from ConvNet
static void	AddVectors (vector< float > &a, vector< float > &b)

Protected Attributes inherited from ConvNet
config::Model	model_

vector< Layer * >	layers_
	The model protobuf config. More...

vector< Layer * >	data_layers_
	The layers in the network. More...

vector< Layer * >	input_layers_
	Layers which have data associated with them. More...

vector< Layer * >	output_layers_
	Input layers. More...

vector< Edge * >	edges_
	Output layers. More...

int	max_iter_
	The edges in the network. More...

int	batch_size_

int	current_iter_

int	lr_reduce_counter_

DataHandler *	train_dataset_

DataHandler *	val_dataset_

string	checkpoint_dir_

string	output_file_

string	model_name_

ImageDisplayer	displayer_

string	model_filename_

string	timestamp_

string	log_file_

string	val_log_file_

Member Function Documentation

void MultiGPUConvNet::Bprop ( bool update_weights )

protectedvirtual

Backpropagate through the network and update weights.

Reimplemented from ConvNet.

void MultiGPUConvNet::Fprop ( bool train )

protectedvirtual

Forward propagate through the network.

Parameters

train If true, this forward prop is being done during training, otherwise during test/validation. Used for determining whether to use drop units stochastcially or use all of them.

Reimplemented from ConvNet.

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

src/multigpu_convnet.h
src/multigpu_convnet.cc

Public Member Functions

Protected Member Functions

Additional Inherited Members

Member Function Documentation