From 4442d2af09e146ae328c39294c7e153d1ec14d26 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?C=C3=A9dric=20Traizet?= <traizetc@cesbio.cnes.fr>
Date: Wed, 24 May 2017 16:54:29 +0200
Subject: [PATCH] added new parameter tied/untied for autoencoders
---
include/AutoencoderModel.txx | 12 +---
.../ImageDimensionalityReductionFilter.txx | 2 +-
include/SOMModel.txx | 47 ++++++----------
include/cbLearningApplicationBaseDR.h | 4 ++
include/cbLearningApplicationBaseDR.txx | 2 +-
include/cbTrainAutoencoder.txx | 55 +++++++++++++++++--
include/cbTrainSOM.txx | 18 +-----
7 files changed, 75 insertions(+), 65 deletions(-)
diff --git a/include/AutoencoderModel.txx b/include/AutoencoderModel.txx
index 7f2b451647..dad8324514 100644
--- a/include/AutoencoderModel.txx
+++ b/include/AutoencoderModel.txx
@@ -57,7 +57,6 @@ void AutoencoderModel<TInputValue,AutoencoderType>::Train()
optimizer.step(error);
std::cout<<i<<" "<<optimizer.solution().value<<std::endl;
}
- //std::cout<<optimizer.solution().value<<std::endl;
m_net.setParameterVector(optimizer.solution().point);
@@ -111,7 +110,6 @@ void AutoencoderModel<TInputValue,AutoencoderType>::Load(const std::string & fil
m_net.read(ia);
ifs.close();
m_NumberOfHiddenNeurons = m_net.numberOfHiddenNeurons();
- //this->m_Size = m_NumberOfHiddenNeurons;
}
@@ -119,6 +117,7 @@ template <class TInputValue, class AutoencoderType>
typename AutoencoderModel<TInputValue,AutoencoderType>::TargetSampleType
AutoencoderModel<TInputValue,AutoencoderType>::DoPredict(const InputSampleType & value) const
{
+ std::cout << "SINGLE PIXEL " ;
shark::RealVector samples(value.Size());
for(size_t i = 0; i < value.Size();i++)
{
@@ -157,17 +156,10 @@ void AutoencoderModel<TInputValue,AutoencoderType>
for(unsigned int a = 0; a < m_NumberOfHiddenNeurons; ++a){
target[a]=p[a];
-
- //target.SetElement(a,p[a]);
-
-
}
- //std::cout << p << std::endl;
targets->SetMeasurementVector(id,target);
- ++id;
-
+ ++id;
}
-
}
diff --git a/include/ImageDimensionalityReductionFilter.txx b/include/ImageDimensionalityReductionFilter.txx
index a072b48fce..3dbfd16bfe 100644
--- a/include/ImageDimensionalityReductionFilter.txx
+++ b/include/ImageDimensionalityReductionFilter.txx
@@ -40,7 +40,7 @@ ImageDimensionalityReductionFilter<TInputImage, TOutputImage, TMaskImage>
this->SetNthOutput(0,TOutputImage::New());
this->SetNthOutput(1,ConfidenceImageType::New());
m_UseConfidenceMap = false;
- m_BatchMode = false;
+ m_BatchMode = true;
}
template <class TInputImage, class TOutputImage, class TMaskImage>
diff --git a/include/SOMModel.txx b/include/SOMModel.txx
index d6c9f02c45..13765ccbfb 100644
--- a/include/SOMModel.txx
+++ b/include/SOMModel.txx
@@ -96,45 +96,32 @@ std::istream & binary_read(std::istream& stream, T& value){
template <class TInputValue, unsigned int MapDimension>
void SOMModel<TInputValue, MapDimension>::Save(const std::string & filename, const std::string & name)
{
-
-//Ecriture
- auto kwl = m_SOMMap->GetImageKeywordlist();
- kwl.AddKey("MachineLearningModelType", "SOM"+std::to_string(MapDimension));
- m_SOMMap->SetImageKeywordList(kwl);
- auto writer = otb::ImageFileWriter<MapType>::New();
- writer->SetInput(m_SOMMap);
- writer->SetFileName(filename);
- writer->Update();
-
-
- // test text
- itk::ImageRegionConstIterator<MapType> inputIterator(m_SOMMap,m_SOMMap->GetLargestPossibleRegion());
- inputIterator.GoToBegin();
- std::ofstream ofs(filename+"2", std::ios::binary);
- binary_write_string(ofs,"som");
- binary_write(ofs,static_cast<int>(MapDimension));
- SizeType size = m_SOMMap->GetLargestPossibleRegion().GetSize() ;
- for (size_t i=0;i<MapDimension;i++){
+ itk::ImageRegionConstIterator<MapType> inputIterator(m_SOMMap,m_SOMMap->GetLargestPossibleRegion());
+ inputIterator.GoToBegin();
+ std::ofstream ofs(filename, std::ios::binary);
+ binary_write_string(ofs,"som");
+ binary_write(ofs,static_cast<int>(MapDimension));
+ SizeType size = m_SOMMap->GetLargestPossibleRegion().GetSize() ;
+ for (size_t i=0;i<MapDimension;i++){
binary_write(ofs,size[i]);
- }
-
- binary_write(ofs,inputIterator.Get().GetNumberOfElements());
- while(!inputIterator.IsAtEnd()){
- InputSampleType vect = inputIterator.Get();
- for (size_t i=0;i<vect.GetNumberOfElements();i++){
- binary_write(ofs,vect[i]);
}
+
+ binary_write(ofs,inputIterator.Get().GetNumberOfElements());
+ while(!inputIterator.IsAtEnd()){
+ InputSampleType vect = inputIterator.Get();
+ for (size_t i=0;i<vect.GetNumberOfElements();i++){
+ binary_write(ofs,vect[i]);
+ }
++inputIterator;
- }
- ofs.close();
-
+ }
+ ofs.close();
}
template <class TInputValue, unsigned int MapDimension>
void SOMModel<TInputValue, MapDimension>::Load(const std::string & filename, const std::string & name)
{
- std::ifstream ifs(filename+"2", std::ios::binary);
+ std::ifstream ifs(filename, std::ios::binary);
/** Read the model key (should be som) */
char s[]=" ";
diff --git a/include/cbLearningApplicationBaseDR.h b/include/cbLearningApplicationBaseDR.h
index 328373b24f..a45024aa47 100644
--- a/include/cbLearningApplicationBaseDR.h
+++ b/include/cbLearningApplicationBaseDR.h
@@ -136,9 +136,13 @@ private:
void InitAutoencoderParams();
void InitPCAParams();
void InitSOMParams();
+
+ void BeforeTrainAutoencoder(typename ListSampleType::Pointer trainingListSample, std::string modelPath);
template <class autoencoderchoice>
void TrainAutoencoder(typename ListSampleType::Pointer trainingListSample, std::string modelPath);
+
void TrainPCA(typename ListSampleType::Pointer trainingListSample, std::string modelPath);
+
template <class somchoice>
void TrainSOM(typename ListSampleType::Pointer trainingListSample, std::string modelPath);
void BeforeTrainSOM(typename ListSampleType::Pointer trainingListSample, std::string modelPath);
diff --git a/include/cbLearningApplicationBaseDR.txx b/include/cbLearningApplicationBaseDR.txx
index 916d7d6f82..5603cca4f6 100644
--- a/include/cbLearningApplicationBaseDR.txx
+++ b/include/cbLearningApplicationBaseDR.txx
@@ -83,7 +83,7 @@ cbLearningApplicationBaseDR<TInputValue,TOutputValue>
if(modelName == "autoencoder")
{
#ifdef OTB_USE_SHARK
- TrainAutoencoder<AutoencoderModelType>(trainingListSample,modelPath);
+ BeforeTrainAutoencoder(trainingListSample,modelPath);
#else
otbAppLogFATAL("Module SharkLearning is not installed. You should consider turning OTB_USE_SHARK on during cmake configuration.");
#endif
diff --git a/include/cbTrainAutoencoder.txx b/include/cbTrainAutoencoder.txx
index feed6cf75e..82e21d17b6 100644
--- a/include/cbTrainAutoencoder.txx
+++ b/include/cbTrainAutoencoder.txx
@@ -21,6 +21,21 @@ cbLearningApplicationBaseDR<TInputValue,TOutputValue>
SetParameterDescription("model.autoencoder",
"This group of parameters allows setting Shark autoencoder parameters. "
);
+
+
+ //Tied Autoencoder
+ AddParameter(ParameterType_Choice, "model.autoencoder.istied",
+ "tied weighth <tied/untied>");
+ SetParameterDescription(
+ "model.autoencoder.istied",
+ "Parameter that determine if the weights are tied or not <tied/untied>");
+
+
+ AddChoice("model.autoencoder.istied.yes","Tied weigths");
+ AddChoice("model.autoencoder.istied.no","Untied weights");
+
+
+
//Number Of Iterations
AddParameter(ParameterType_Int, "model.autoencoder.nbiter",
"Maximum number of iterations during training");
@@ -38,13 +53,41 @@ cbLearningApplicationBaseDR<TInputValue,TOutputValue>
"model.autoencoder.nbneuron",
"The number of neurons in the hidden layer.");
- //normalization
- AddParameter(ParameterType_Float, "model.autoencoder.normalizer", "Strength of the normalization");
- SetParameterFloat("model.autoencoder.normalizer",0, false);
- SetParameterDescription("model.autoencoder.normalizer",
- "Strength of the L2 normalization used during training");
+ //Regularization
+ AddParameter(ParameterType_Float, "model.autoencoder.regularization", "Strength of the regularization");
+ SetParameterFloat("model.autoencoder.regularization",0, false);
+ SetParameterDescription("model.autoencoder.regularization",
+ "Strength of the L2 regularization used during training");
+}
+
+
+template <class TInputValue, class TOutputValue>
+void
+cbLearningApplicationBaseDR<TInputValue,TOutputValue>
+::BeforeTrainAutoencoder(typename ListSampleType::Pointer trainingListSample,
+ std::string modelPath)
+{
+ std::string TiedWeigth = GetParameterString("model.autoencoder.istied");
+ std::cout << TiedWeigth << std::endl;
+
+ if(TiedWeigth == "no")
+ {
+ TrainAutoencoder<AutoencoderModelType>(trainingListSample,modelPath);
+ }
+
+ if(TiedWeigth == "yes")
+ {
+ TrainAutoencoder<TiedAutoencoderModelType>(trainingListSample,modelPath);
+ }
+
+ if(TiedWeigth != "yes" && TiedWeigth != "no")
+ {
+ std::cerr << "istied : invalid choice <yes/no>" << std::endl;
+ }
}
+
+
template <class TInputValue, class TOutputValue>
template <typename autoencoderchoice>
void cbLearningApplicationBaseDR<TInputValue,TOutputValue>
@@ -53,7 +96,7 @@ void cbLearningApplicationBaseDR<TInputValue,TOutputValue>
typename autoencoderchoice::Pointer dimredTrainer = autoencoderchoice::New();
dimredTrainer->SetNumberOfHiddenNeurons(GetParameterInt("model.autoencoder.nbneuron"));
dimredTrainer->SetNumberOfIterations(GetParameterInt("model.autoencoder.nbiter"));
- dimredTrainer->SetRegularization(GetParameterFloat("model.autoencoder.normalizer"));
+ dimredTrainer->SetRegularization(GetParameterFloat("model.autoencoder.regularization"));
dimredTrainer->SetInputListSample(trainingListSample);
dimredTrainer->Train();
dimredTrainer->Save(modelPath);
diff --git a/include/cbTrainSOM.txx b/include/cbTrainSOM.txx
index 539504b92b..9ac28dd906 100644
--- a/include/cbTrainSOM.txx
+++ b/include/cbTrainSOM.txx
@@ -18,22 +18,6 @@ cbLearningApplicationBaseDR<TInputValue,TOutputValue>
SetParameterDescription("model.som",
"This group of parameters allows setting SOM parameters. "
);
-
- AddChoice("model.som3d", "OTB SOM");
- SetParameterDescription("model.som3d",
- "This group of parameters allows setting SOM parameters. "
- );
-
- AddChoice("model.som4d", "OTB SOM");
- SetParameterDescription("model.som4d",
- "This group of parameters allows setting SOM parameters. "
- );
-
- AddChoice("model.som5d", "OTB SOM");
- SetParameterDescription("model.som5d",
- "This group of parameters allows setting SOM parameters. "
- );
-
AddParameter(ParameterType_Int, "model.som.dim","Dimension of the map");
SetParameterDescription("model.som.dim","Dimension of the SOM map.");
@@ -120,7 +104,7 @@ cbLearningApplicationBaseDR<TInputValue,TOutputValue>
}
if(SomDim > 5 || SomDim < 2)
{
- std::cerr << "invalid dimension" << std::endl;
+ std::cerr << "k : invalid dimension" << std::endl;
}
}
--
GitLab