GPUMLib  0.2.2
GPU Machine Learning Library
Public Member Functions | List of all members

Represents a Deep Belief Network (Device - GPU). More...

#include <DBN.h>

Public Member Functions

 DBN (HostArray< int > &layers, HostMatrix< cudafloat > &inputs, cudafloat initialLearningRate, cudafloat momentum=DEFAULT_MOMENTUM, bool useBinaryValuesVisibleReconstruction=false, cudafloat stdWeights=STD_WEIGHTS)
 
RBMGetRBM (int layer)
 
int GetNumberRBMs () const
 
bool Train (int epochs, int cd, int miniBatchSize, cudafloat errorStop=CUDA_VALUE(0.0))
 

Detailed Description

Represents a Deep Belief Network (Device - GPU).

Examples:
DBNapp.cpp.

Definition at line 33 of file DBN.h.

Constructor & Destructor Documentation

DBN ( HostArray< int > &  layers,
HostMatrix< cudafloat > &  inputs,
cudafloat  initialLearningRate,
cudafloat  momentum = DEFAULT_MOMENTUM,
bool  useBinaryValuesVisibleReconstruction = false,
cudafloat  stdWeights = STD_WEIGHTS 
)
inline

Constructs a Deep Belief Network that can be trained using the CPU (Host).

Parameters
layersNumber of units in each layer.
inputsInputs of the training dataset. Each row of the matrix should contain a pattern (sample) and each column an input.
initialLearningRateInitial learning rate.
momentumMomentum
useBinaryValuesVisibleReconstructionUse binary values for the visibible layer reconstruction
stdWeightsDefines the maximum and minimum value for the weights. The weights will be initialized with a random number between -stdWeights and stdWeights.

Definition at line 45 of file DBN.h.

Member Function Documentation

int GetNumberRBMs ( ) const
inline

Gets the number of RBMs.

Returns
The number of RBMs that compose the DBN.
Examples:
DBNapp.cpp.

Definition at line 84 of file DBN.h.

RBM* GetRBM ( int  layer)
inline

Get an RBM that is part of the DBN.

Parameters
layerLayer to obtain.
Returns
The RBM corresponding to layer specified.
Examples:
DBNapp.cpp.

Definition at line 77 of file DBN.h.

bool Train ( int  epochs,
int  cd,
int  miniBatchSize,
cudafloat  errorStop = CUDA_VALUE(0.0) 
)
inline

Train the DBN (train each RBM).

Parameters
epochsMaximum number of epochs that each RBM should be trained.
cdDefine the value of k in CD-k.
miniBatchSizeMini Batch Size.
errorStopStop the training in each RBM when the Mean Square Error (MSE) is inferior to this value.
Returns
True if successful. False otherwise.

Definition at line 94 of file DBN.h.


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