/**
* Create the ConvLayer object using the specified number of input maps,
* output maps, filter size, stride and padding parameter.
*
* @param inMaps The number of input maps.
* @param outMaps The number of output maps.
* @param wfilter Width of the filter/kernel.
* @param wfilter Height of the filter/kernel.
* @param xStride Stride of filter application in the x direction.
* @param yStride Stride of filter application in the y direction.
* @param wPad Spatial padding width of the input.
* @param hPad Spatial padding height of the input.
* @param WeightInitRule The weight initialization rule used to initialize the
* weight matrix.
*/
ConvLayer(const size_t inMaps,
const size_t outMaps,
const size_t wfilter,
const size_t hfilter,
const size_t xStride = 1,
const size_t yStride = 1,
const size_t wPad = 0,
const size_t hPad = 0,
WeightInitRule weightInitRule = WeightInitRule()) :
wfilter(wfilter),
hfilter(hfilter),
inMaps(inMaps),
outMaps(outMaps),
xStride(xStride),
yStride(yStride),
wPad(wPad),
hPad(hPad),
optimizer(new OptimizerType<ConvLayer<OptimizerType,
WeightInitRule,
ForwardConvolutionRule,
BackwardConvolutionRule,
GradientConvolutionRule,
InputDataType,
OutputDataType>,
OutputDataType>(*this)),
ownsOptimizer(true)
{
weightInitRule.Initialize(weights, wfilter, hfilter, inMaps * outMaps);
}
/**
* Create the FullConnection object using the specified input layer, output
* layer, optimizer and weight initialize rule.
*
* @param InputLayerType The input layer which is connected with the output
* layer.
* @param OutputLayerType The output layer which is connected with the input
* layer.
* @param OptimizerType The optimizer used to update the weight matrix.
* @param WeightInitRule The weight initialize rule used to initialize the
* weight matrix.
*/
FullConnection(InputLayerType& inputLayer,
OutputLayerType& outputLayer,
OptimizerType& optimizer,
WeightInitRule weightInitRule = WeightInitRule()) :
inputLayer(inputLayer), outputLayer(outputLayer), optimizer(optimizer)
{
weightInitRule.Initialize(weights, outputLayer.InputSize(),
inputLayer.OutputSize());
}
/**
* Create the SelfConnection object using the specified input layer, output
* layer, optimizer and weight initialize rule.
*
* @param InputLayerType The input layer which is connected with the output
* layer.
* @param OutputLayerType The output layer which is connected with the input
* layer.
* @param OptimizerType The optimizer used to update the weight matrix.
* @param WeightInitRule The weight initialize rule used to initialize the
* weight matrix.
*/
SelfConnection(InputLayerType& inputLayer,
OutputLayerType& outputLayer,
OptimizerType& optimizer,
WeightInitRule weightInitRule = WeightInitRule()) :
inputLayer(inputLayer),
outputLayer(outputLayer),
optimizer(optimizer),
connection(1 - arma::eye<MatType>(inputLayer.OutputSize(),
inputLayer.OutputSize()))
{
weightInitRule.Initialize(weights, outputLayer.InputSize(),
inputLayer.OutputSize());
}
/**
* Create the BiasLayer object using the specified number of units and bias
* parameter.
*
* @param outSize The number of output units.
* @param bias The bias value.
* @param WeightInitRule The weight initialization rule used to initialize the
* weight matrix.
*/
BiasLayer(const size_t outSize,
const double bias = 1,
WeightInitRule weightInitRule = WeightInitRule()) :
outSize(outSize),
bias(bias),
optimizer(new OptimizerType<BiasLayer<OptimizerType,
WeightInitRule,
InputDataType,
OutputDataType>,
InputDataType>(*this)),
ownsOptimizer(true)
{
weightInitRule.Initialize(weights, outSize, 1);
}
/**
* Create the SelfConnection object using the specified input layer, output
* layer, optimizer and weight initialize rule.
*
* @param InputLayerType The input layer which is connected with the output
* layer.
* @param OutputLayerType The output layer which is connected with the input
* layer.
* @param OptimizerType The optimizer used to update the weight matrix.
* @param WeightInitRule The weight initialize rule used to initialize the
* weight matrix.
*/
SelfConnection(InputLayerType& inputLayer,
OutputLayerType& outputLayer,
WeightInitRule weightInitRule = WeightInitRule()) :
inputLayer(inputLayer),
outputLayer(outputLayer),
optimizer(new OptimizerType<SelfConnection<InputLayerType,
OutputLayerType,
OptimizerType,
WeightInitRule,
MatType,
VecType>, MatType>(*this)),
ownsOptimizer(true),
connection(1 - arma::eye<MatType>(inputLayer.OutputSize(),
inputLayer.OutputSize()))
{
weightInitRule.Initialize(weights, outputLayer.InputSize(),
inputLayer.OutputSize());
}
/**
* Create the ConvConnection object using the specified input layer, output
* layer, filter size and weight initialization rule.
*
* @param InputLayerType The input layer which is connected with the output
* layer.
* @param OutputLayerType The output layer which is connected with the input
* layer.
* @param filterSize the size of the filter.
* @param WeightInitRule The weights initialization rule used to initialize
* the weights matrix.
*/
ConvConnection(InputLayerType& inputLayer,
OutputLayerType& outputLayer,
const size_t filterSize,
WeightInitRule weightInitRule = WeightInitRule()) :
inputLayer(inputLayer),
outputLayer(outputLayer),
optimizer(new OptimizerType<ConvConnection<InputLayerType,
OutputLayerType,
OptimizerType,
WeightInitRule,
ForwardConvolutionRule,
BackwardConvolutionRule,
GradientConvolutionRule,
DataType>, DataType>(*this)),
ownsOptimizer(true)
{
weightInitRule.Initialize(weights, filterSize, filterSize,
inputLayer.OutputMaps() * outputLayer.OutputMaps());
}
/**
* Create the ConvConnection object using the specified input layer, output
* layer, filter size, optimizer and weight initialization rule.
*
* @param InputLayerType The input layer which is connected with the output
* layer.
* @param OutputLayerType The output layer which is connected with the input
* layer.
* @param filterSize the size of the filter.
* @param OptimizerType The optimizer used to update the weight matrix.
* @param WeightInitRule The weights initialization rule used to initialize
* the weights matrix.
*/
ConvConnection(InputLayerType& inputLayer,
OutputLayerType& outputLayer,
const size_t filterSize,
OptimizerType<ConvConnection<InputLayerType,
OutputLayerType,
OptimizerType,
WeightInitRule,
ForwardConvolutionRule,
BackwardConvolutionRule,
GradientConvolutionRule,
DataType>, DataType>& optimizer,
WeightInitRule weightInitRule = WeightInitRule()) :
inputLayer(inputLayer),
outputLayer(outputLayer),
optimizer(&optimizer),
ownsOptimizer(false)
{
weightInitRule.Initialize(weights, filterSize, filterSize,
outputLayer.LayerSlices());
}
/**
* Create the LSTMLayer object using the specified parameters.
*
* @param layerSize The number of memory cells.
* @param layerSize The length of the input sequence.
* @param peepholes The flag used to indicate if peephole connections should
* be used (Default: true).
* @param WeightInitRule The weight initialize rule used to initialize the
* peephole connection matrix.
*/
LSTMLayer(const size_t layerSize,
const size_t seqLen = 1,
const bool peepholes = false,
WeightInitRule weightInitRule = WeightInitRule()) :
inputActivations(arma::zeros<VecType>(layerSize * 4)),
layerSize(layerSize),
seqLen(seqLen),
inGate(arma::zeros<MatType>(layerSize, seqLen)),
inGateAct(arma::zeros<MatType>(layerSize, seqLen)),
inGateError(arma::zeros<MatType>(layerSize, seqLen)),
outGate(arma::zeros<MatType>(layerSize, seqLen)),
outGateAct(arma::zeros<MatType>(layerSize, seqLen)),
outGateError(arma::zeros<MatType>(layerSize, seqLen)),
forgetGate(arma::zeros<MatType>(layerSize, seqLen)),
forgetGateAct(arma::zeros<MatType>(layerSize, seqLen)),
forgetGateError(arma::zeros<MatType>(layerSize, seqLen)),
state(arma::zeros<MatType>(layerSize, seqLen)),
stateError(arma::zeros<MatType>(layerSize, seqLen)),
cellAct(arma::zeros<MatType>(layerSize, seqLen)),
offset(0),
peepholes(peepholes)
{
if (peepholes)
{
weightInitRule.Initialize(inGatePeepholeWeights, layerSize, 1);
inGatePeepholeDerivatives = arma::zeros<VecType>(layerSize);
inGatePeepholeOptimizer = std::unique_ptr<OptimizerType>(
new OptimizerType(1, layerSize));
weightInitRule.Initialize(forgetGatePeepholeWeights, layerSize, 1);
forgetGatePeepholeDerivatives = arma::zeros<VecType>(layerSize);
forgetGatePeepholeOptimizer = std::unique_ptr<OptimizerType>(
new OptimizerType(1, layerSize));
weightInitRule.Initialize(outGatePeepholeWeights, layerSize, 1);
outGatePeepholeDerivatives = arma::zeros<VecType>(layerSize);
outGatePeepholeOptimizer = std::unique_ptr<OptimizerType>(
new OptimizerType(1, layerSize));
}
}