From b18d93a8b78d96ac192f0de87c5466d9a9296030 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?C=C3=A9dric=20Traizet?= <cedric.traizet@c-s.fr>
Date: Wed, 17 Apr 2019 17:26:30 +0200
Subject: [PATCH] ENH: remove normalization from Shark Kmeans machine learning
model
---
.../otbSharkKMeansMachineLearningModel.h | 8 -------
.../otbSharkKMeansMachineLearningModel.hxx | 21 +------------------
2 files changed, 1 insertion(+), 28 deletions(-)
diff --git a/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.h b/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.h
index 700a58eacd..d42e748194 100644
--- a/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.h
+++ b/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.h
@@ -125,10 +125,6 @@ public:
/** Set the number of class for the kMeans algorithm.*/
itkSetMacro( K, unsigned );
- /** If true, normalized input data sample list */
- itkGetMacro( Normalized, bool );
- itkSetMacro( Normalized, bool );
-
/** Initialize the centroids for the kmeans algorithm */
void SetCentroidsFromData(const shark::Data<shark::RealVector> & data)
{
@@ -150,9 +146,6 @@ protected:
virtual void DoPredictBatch(const InputListSampleType *, const unsigned int &startIndex, const unsigned int &size,
TargetListSampleType *, ConfidenceListSampleType * = nullptr, ProbaListSampleType * = nullptr) const override;
- template<typename DataType>
- DataType NormalizeData(const DataType &data) const;
-
template<typename DataType>
shark::Normalizer<> TrainNormalizer(const DataType &data) const;
@@ -164,7 +157,6 @@ private:
void operator=(const Self &) = delete;
// Parameters set by the user
- bool m_Normalized;
unsigned int m_K;
unsigned int m_MaximumNumberOfIterations;
bool m_CanRead;
diff --git a/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.hxx b/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.hxx
index 1d27ce519c..ff95c9501a 100644
--- a/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.hxx
+++ b/Modules/Learning/Unsupervised/include/otbSharkKMeansMachineLearningModel.hxx
@@ -52,7 +52,7 @@ namespace otb
template<class TInputValue, class TOutputValue>
SharkKMeansMachineLearningModel<TInputValue, TOutputValue>
::SharkKMeansMachineLearningModel() :
- m_Normalized( false ), m_K(2), m_MaximumNumberOfIterations( 10 )
+ m_K(2), m_MaximumNumberOfIterations( 10 )
{
// Default set HardClusteringModel
this->m_ConfidenceIndex = true;
@@ -77,30 +77,11 @@ SharkKMeansMachineLearningModel<TInputValue, TOutputValue>
otb::Shark::ListSampleToSharkVector( this->GetInputListSample(), vector_data );
shark::Data<shark::RealVector> data = shark::createDataFromRange( vector_data );
- // Normalized input value if necessary
- if( m_Normalized )
- {
- auto normalizer = TrainNormalizer(data);
- data = normalizer(data);
- }
-
// Use a Hard Clustering Model for classification
shark::kMeans( data, m_K, m_Centroids, m_MaximumNumberOfIterations );
m_ClusteringModel = boost::make_shared<ClusteringModelType>( &m_Centroids );
}
-template<class TInputValue, class TOutputValue>
-template<typename DataType>
-DataType
-SharkKMeansMachineLearningModel<TInputValue, TOutputValue>
-::NormalizeData(const DataType &data) const
-{
- shark::Normalizer<> normalizer;
- shark::NormalizeComponentsUnitVariance<> normalizingTrainer( true );//zero mean
- normalizingTrainer.train( normalizer, data );
- return normalizer( data );
-}
-
template<class TInputValue, class TOutputValue>
template<typename DataType>
shark::Normalizer<>
--
GitLab