libinsighttoolkit4-dev 4.9.0-4ubuntu1 / usr / include / ITK-4.9 / itkRBFLayer.hxx

/*=========================================================================
 *
 *  Copyright Insight Software Consortium
 *
 *  Licensed under the Apache License, Version 2.0 (the "License");
 *  you may not use this file except in compliance with the License.
 *  You may obtain a copy of the License at
 *
 *         http://www.apache.org/licenses/LICENSE-2.0.txt
 *
 *  Unless required by applicable law or agreed to in writing, software
 *  distributed under the License is distributed on an "AS IS" BASIS,
 *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 *  See the License for the specific language governing permissions and
 *  limitations under the License.
 *
 *=========================================================================*/
#ifndef itkRBFLayer_hxx
#define itkRBFLayer_hxx

#include "itkRBFLayer.h"
#include "itkGaussianRadialBasisFunction.h"

namespace itk
{
namespace Statistics
{
template<typename TMeasurementVector, typename TTargetVector>
RBFLayer<TMeasurementVector,TTargetVector>
::RBFLayer()
{
  m_Bias = 1;
  m_NumClasses = 0;
  typedef GaussianRadialBasisFunction<ValueType> GRBFType;
  m_RBF=GRBFType::New();
  m_DistanceMetric = DistanceMetricType::New();
  //  TMeasurementVector origin;

  //  m_DistanceMetric->SetMeasurementVectorSize(origin.Size());
  m_RBF_Dim = 0;
  //
}


template<typename TMeasurementVector, typename TTargetVector>
RBFLayer<TMeasurementVector,TTargetVector>
::~RBFLayer()
{
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetRBF(RBFType* f)
{
  m_RBF = f;
  this->Modified();
}
template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetRBF_Dim(unsigned int dim)
{
  m_RBF_Dim=dim;
  m_DistanceMetric->SetMeasurementVectorSize(m_RBF_Dim);
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetNumberOfNodes(unsigned int c)
{
  //TMeasurementVector sampleinputvector;
  //m_RBF_Dim= sampleinputvector.Size();
  Superclass::SetNumberOfNodes(c);
  this->m_NodeInputValues.set_size(m_RBF_Dim); //c);
  this->m_NodeOutputValues.set_size(c);
  m_InputErrorValues.set_size(c);
  m_OutputErrorValues.set_size(c);

  if(this->GetLayerTypeCode() != Self::OUTPUTLAYER)
    {
    //TMeasurementVector temp;
    InternalVectorType temp(m_RBF_Dim);
    for(unsigned int i=0; i<c; i++)
      {
      m_Centers.push_back(temp);
      }
    this->m_NodeOutputValues.set_size(c);
    m_Radii.SetSize(c);
    m_Radii.fill(1.0);
    }
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetInputValue(unsigned int i, ValueType value)
{
  this->m_NodeInputValues[i] = value;
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::GetInputValue(unsigned int i) const
{
  return m_NodeInputValues[i];
}
template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetOutputValue(unsigned int i, ValueType value)
{
  m_NodeOutputValues(i) = value;
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::GetOutputValue(unsigned int i) const
{
  return m_NodeOutputValues(i);
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetOutputVector(TMeasurementVector value)
{
  m_NodeOutputValues = value.GetVnlVector();
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType *
RBFLayer<TMeasurementVector,TTargetVector>
::GetOutputVector()
{
  return m_NodeOutputValues.data_block();
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetRadii(ValueType c,unsigned int i)
{
  m_Radii.SetElement(i,c);
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::GetRadii(unsigned int i) const
{
  return m_Radii.GetElement(i);
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetCenter(TMeasurementVector c,unsigned int i)
{
  InternalVectorType temp(c.Size());
  for(unsigned int j=0; j<c.Size(); j++)
    {
    temp[j]=c[j];
    }
  m_Centers[i]=temp; //c;
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::InternalVectorType
RBFLayer<TMeasurementVector,TTargetVector>
::GetCenter(unsigned int i) const
{
  if(m_Centers.size() != 0)
    {
    return m_Centers[i];
    }
  else
    {
    return 0;
    }
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::GetInputErrorValue(unsigned int n) const
{
  return m_InputErrorValues[n];
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType *
RBFLayer<TMeasurementVector,TTargetVector>
::GetInputErrorVector()
{
  return m_InputErrorValues.data_block();
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetInputErrorValue(ValueType v, unsigned int i)
{
  m_InputErrorValues[i] = v;
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::ForwardPropagate()
{
  typename WeightSetInterfaceType::Pointer inputweightset;
  typename InputFunctionInterfaceType::Pointer inputfunction;
  if(this->GetLayerTypeCode() == Self::OUTPUTLAYER)
    {
    typename TransferFunctionInterfaceType::Pointer transferfunction;

    inputfunction = this->GetModifiableNodeInputFunction();
    transferfunction = this->GetModifiableActivationFunction();
    inputweightset = this->GetModifiableInputWeightSet();
    ValueType * inputvalues = inputweightset->GetOutputValues();

    const int rows = this->m_NumberOfNodes;
    const int cols = this->m_InputWeightSet->GetNumberOfInputNodes();
    vnl_matrix<ValueType> inputmatrix;
    inputmatrix.set_size(rows, cols);
    inputmatrix.copy_in(inputvalues);

    inputfunction->SetSize(cols); //include bias
    for (int j = 0; j < rows; j++)
      {
      vnl_vector<ValueType> temp_vnl;
      temp_vnl.set_size(inputmatrix.cols());
      temp_vnl=inputmatrix.get_row(j);
      m_NodeInputValues.put(j, inputfunction->Evaluate(temp_vnl.data_block()));
      m_NodeOutputValues.put(j, transferfunction->Evaluate(m_NodeInputValues[j]));
      }
    }
  else
    {
    inputweightset = this->GetModifiableInputWeightSet();
    inputfunction = this->GetModifiableNodeInputFunction();

    vnl_vector<ValueType> temp;
    ValueType * inputvalues = inputweightset->GetInputValues();

    int cols = this->m_InputWeightSet->GetNumberOfInputNodes();
    vnl_matrix<ValueType> inputmatrix;
    inputmatrix.set_size(1, cols-1);
    inputmatrix.copy_in(inputvalues);
    inputfunction->SetSize(cols-1); //include bias
    m_NodeInputValues = inputmatrix.get_row(0);
    ValueType * cdeltavalues = inputweightset->GetTotalDeltaValues();
    vnl_matrix<ValueType> center_increment(cdeltavalues,inputweightset->GetNumberOfOutputNodes(),
                                           inputweightset->GetNumberOfInputNodes());
    vnl_vector<ValueType> width_increment;
    width_increment.set_size(inputweightset->GetNumberOfOutputNodes());
    width_increment.fill(0);
    width_increment= center_increment.get_column(inputweightset->GetNumberOfInputNodes()-1);
    ValueType temp_radius;
    InternalVectorType temp_center;
    temp_center.SetSize(m_RBF_Dim);
    //TMeasurementVector tempvector1;
    //TMeasurementVector tempvector2;
    //TMeasurementVector tempcenter;
    InternalVectorType tempvector1(m_RBF_Dim);
    InternalVectorType tempvector2(m_RBF_Dim);
    InternalVectorType tempcenter(m_RBF_Dim);

    for (unsigned int i = 0; i < m_NumClasses; i++)
      {
      tempcenter = m_Centers[i];
      for(unsigned int j=0;j<m_RBF_Dim;j++)
        {
        ValueType val =tempcenter[j];
        val += center_increment[i][j];
        tempcenter[j]=val;
        }

      m_Centers[i]=tempcenter;
      temp_radius = m_Radii.GetElement(i);
      temp_radius += width_increment[i];
      m_Radii.SetElement(i,temp_radius);
      InternalVectorType array1(m_NodeInputValues.size());

      array1= m_NodeInputValues;

      for(unsigned int j=0; j<tempvector1.size(); j++)
        tempvector1[j]=m_NodeInputValues[j];

      //tempvector1.Set_vnl_vector(m_NodeInputValues);
      tempvector2=m_Centers[i];
      tempcenter= m_Centers[i];
      //double dt= m_DistanceMetric->Evaluate(tempvector1,tempvector2);
      //std::cout<<"Euclidean in layer ="<<dt<<std::endl;
      m_RBF->SetRadius(m_Radii.GetElement(i));
      InternalVectorType temp_array(m_RBF_Dim);
      NodeVectorType temp_vector=  m_Centers[i];
      for(unsigned int ii=0; ii<m_RBF_Dim; ii++)
        temp_array.SetElement(ii,temp_vector[ii]);
      m_RBF->SetCenter(temp_array);
      m_NodeOutputValues.put(i,m_RBF->Evaluate(m_DistanceMetric->Evaluate(tempvector1,tempvector2)));
      }
    }
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetDistanceMetric(DistanceMetricType* f)
{
  m_DistanceMetric=f;
  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::ForwardPropagate(TMeasurementVector samplevector)
{
  typename TransferFunctionInterfaceType::Pointer transferfunction = this->GetModifiableActivationFunction();

  for (unsigned int i = 0; i < samplevector.Size(); i++)
    {
    samplevector[i] = transferfunction->Evaluate(samplevector[i]);
    m_NodeOutputValues.put(i, samplevector[i]);
    }
}


template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::SetOutputErrorValues(TTargetVector errors)
{

  for(unsigned int i=0; i<errors.Size(); i++)
    m_OutputErrorValues[i] = errors[i];

  this->Modified();
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::GetOutputErrorValue(unsigned int i) const
{
  return m_OutputErrorValues[i];
}


template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::BackwardPropagate()
{
  const unsigned int num_nodes = this->GetNumberOfNodes();

  typename Superclass::WeightSetType::Pointer outputweightset = Superclass::GetModifiableOutputWeightSet();
  typename Superclass::WeightSetType::Pointer inputweightset = Superclass::GetModifiableInputWeightSet();

  vnl_vector<ValueType> OutputLayerInput(outputweightset->GetInputValues(),num_nodes);


  ValueType * deltavalues = outputweightset->GetDeltaValues();
  ValueType * weightvalues = outputweightset->GetWeightValues();

  const unsigned int cols = num_nodes;
  const unsigned int rows = outputweightset->GetNumberOfOutputNodes();

  vnl_matrix<ValueType> weightmatrix(weightvalues, rows, cols);

  vnl_matrix<ValueType> deltamatrix(deltavalues, rows, cols);
  vnl_vector<ValueType> deltaww;
  deltaww.set_size(cols);
  deltaww.fill(0);
  /*
        TMeasurementVector tempvector1;
        TMeasurementVector tempvector2;*/

  InternalVectorType tempvector1(m_RBF_Dim);
  InternalVectorType tempvector2(m_RBF_Dim);

  for(unsigned int c1=0; c1<rows; c1++)
    {
    for(unsigned int c2=0; c2<cols; c2++)
      {
      deltamatrix[c1][c2]=deltamatrix[c1][c2]/OutputLayerInput[c2];
      }
    }
  for (unsigned int i = 0; i < cols; i++)
    {
    deltaww[i] = dot_product(deltamatrix.get_column(i),
                             weightmatrix.get_column(i));
    }

  //compute gradient for centers
  InternalVectorType array1(m_NodeInputValues.size());
  array1= m_NodeInputValues;
  vnl_matrix<ValueType> DW_temp(inputweightset->GetNumberOfOutputNodes(),
                                inputweightset->GetNumberOfInputNodes());
  DW_temp.fill(0.0);

  for(unsigned int k=0; k<array1.Size(); k++)
    tempvector1[k]=array1[k];

  for(unsigned int k=0; k<num_nodes; k++)
    {
    for (unsigned int i = 0; i < m_RBF_Dim; i++)
      {
      tempvector2=m_Centers[k];
      double dist=m_DistanceMetric->Evaluate(tempvector1,tempvector2);
      m_RBF->SetRadius(m_Radii.GetElement(k));
      NodeVectorType temp_vector=  m_Centers[k];
      InternalVectorType temp_array(m_RBF_Dim);
      for(unsigned int ii=0; ii<m_RBF_Dim; ii++)
        temp_array.SetElement(ii,temp_vector[ii]);
      m_RBF->SetCenter(temp_array);

      DW_temp[k][i]=deltaww[k] * m_RBF->EvaluateDerivative
        (dist,array1,'u',i);
      }
    }

  //compute gradient for widths
  NodeVectorType width_gradient;
  width_gradient.set_size(num_nodes);
  width_gradient.fill(0.0);

  for (unsigned int i=0;i<num_nodes;i++)
    {
    tempvector2=m_Centers[i];
    double dist=m_DistanceMetric->Evaluate(tempvector1,tempvector2);
    width_gradient[i]=deltaww[i] * m_RBF->EvaluateDerivative
      (dist,array1,'s');
    }
  inputweightset->SetDeltaValues(DW_temp.data_block());
  inputweightset->SetDeltaBValues(width_gradient.data_block());
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::Activation(ValueType n)
{
  return this->m_ActivationFunction->Evaluate(n);
}

template<typename TMeasurementVector, typename TTargetVector>
typename RBFLayer<TMeasurementVector,TTargetVector>::ValueType
RBFLayer<TMeasurementVector,TTargetVector>
::DActivation(ValueType n)
{
  return this->m_ActivationFunction->EvaluateDerivative(n);
}

/** Print the object */
template<typename TMeasurementVector, typename TTargetVector>
void
RBFLayer<TMeasurementVector,TTargetVector>
::PrintSelf( std::ostream& os, Indent indent ) const
{
  os << indent << "RBFLayer(" << this << ")" << std::endl;
  os << indent << "m_DistanceMetric = " << m_DistanceMetric << std::endl;
  os << indent << "m_NodeInputValues = " << m_NodeInputValues << std::endl;
  os << indent << "m_NodeOutputValues = " << m_NodeOutputValues << std::endl;
  os << indent << "m_InputErrorValues = " << m_InputErrorValues << std::endl;
  os << indent << "m_OutputErrorValues = " << m_OutputErrorValues << std::endl;
  //os << indent << "m_Centers = " << m_Centers << std::endl;
  os << indent << "m_Radii = " << m_Radii << std::endl;
  os << indent << "m_Bias = " << m_Bias << std::endl;
  os << indent << "m_RBF_Dim = " << m_RBF_Dim << std::endl;
  os << indent << "m_RBF = " << m_RBF << std::endl;
  Superclass::PrintSelf( os, indent );
}

} // end namespace Statistics
} // end namespace itk

#endif