otb::sampleAugmentation Namespace Reference

Classes
struct	NeighborSorter
struct	NeighborType

Typedefs
using	SampleType = std::vector< double >
using	SampleVectorType = std::vector< SampleType >
using	NNIndicesType = std::vector< NeighborType >
using	NNVectorType = std::vector< NNIndicesType >
SampleType	EstimateStds (const SampleVectorType &samples)
void	ReplicateSamples (const SampleVectorType &inSamples, const vcl_size_t nbSamples, SampleVectorType &newSamples)
void	JitterSamples (const SampleVectorType &inSamples, const vcl_size_t nbSamples, SampleVectorType &newSamples, float stdFactor=10, const int seed=std::time(nullptr))
double	ComputeSquareDistance (const SampleType &x, const SampleType &y)
void	FindKNNIndices (const SampleVectorType &inSamples, const vcl_size_t nbNeighbors, NNVectorType &nnVector)
SampleType	SmoteCombine (const SampleType &s1, const SampleType &s2, double position)
void	Smote (const SampleVectorType &inSamples, const vcl_size_t nbSamples, SampleVectorType &newSamples, const int nbNeighbors, const int seed=std::time(nullptr))

Typedef Documentation

NNIndicesType

using otb::sampleAugmentation::NNIndicesType = typedef std::vector<NeighborType>

Estimate standard deviations of the components in one pass using Welford's algorithm

Definition at line 164 of file otbSampleAugmentation.h.

NNVectorType

using otb::sampleAugmentation::NNVectorType = typedef std::vector<NNIndicesType>

Estimate standard deviations of the components in one pass using Welford's algorithm

Definition at line 165 of file otbSampleAugmentation.h.

SampleType

using otb::sampleAugmentation::SampleType = typedef std::vector<double>

Definition at line 41 of file otbSampleAugmentation.h.

SampleVectorType

using otb::sampleAugmentation::SampleVectorType = typedef std::vector<SampleType>

Definition at line 42 of file otbSampleAugmentation.h.

Function Documentation

ComputeSquareDistance()

double otb::sampleAugmentation::ComputeSquareDistance	(	const SampleType &	x,
		const SampleType &	y
	)

Estimate standard deviations of the components in one pass using Welford's algorithm

Definition at line 153 of file otbSampleAugmentation.h.

Referenced by FindKNNIndices().

EstimateStds()

SampleType otb::sampleAugmentation::EstimateStds

(

const SampleVectorType &

samples

)

Estimate standard deviations of the components in one pass using Welford's algorithm

Definition at line 48 of file otbSampleAugmentation.h.

Referenced by JitterSamples().

FindKNNIndices()

void otb::sampleAugmentation::FindKNNIndices	(	const SampleVectorType &	inSamples,
		const vcl_size_t	nbNeighbors,
		NNVectorType &	nnVector
	)

Returns the indices of the nearest neighbors for each input sample

Definition at line 168 of file otbSampleAugmentation.h.

References ComputeSquareDistance().

Referenced by Smote().

JitterSamples()

void otb::sampleAugmentation::JitterSamples	(	const SampleVectorType &	inSamples,
		const vcl_size_t	nbSamples,
		SampleVectorType &	newSamples,
		float	stdFactor = 10,
		const int	seed = std::time(nullptr)
	)

Create new samples by adding noise to existing samples. Gaussian noise is added to randomly selected samples. The standard deviation of the noise added to each component is the same as the one of the input variables divided by stdFactor (defaults to 10). The elements of newSamples are removed before proceeding.

Definition at line 107 of file otbSampleAugmentation.h.

References EstimateStds().

ReplicateSamples()

void otb::sampleAugmentation::ReplicateSamples	(	const SampleVectorType &	inSamples,
		const vcl_size_t	nbSamples,
		SampleVectorType &	newSamples
	)

Create new samples by replicating input samples. We loop through the input samples and add them to the new data set until nbSamples are added. The elements of newSamples are removed before proceeding.

Definition at line 84 of file otbSampleAugmentation.h.

Smote()

void otb::sampleAugmentation::Smote	(	const SampleVectorType &	inSamples,
		const vcl_size_t	nbSamples,
		SampleVectorType &	newSamples,
		const int	nbNeighbors,
		const int	seed = std::time(nullptr)
	)

Create new samples using the SMOTE algorithm Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P., Smote: synthetic minority over-sampling technique, Journal of artificial intelligence research, 16(), 321–357 (2002). http://dx.doi.org/10.1613/jair.953

Definition at line 221 of file otbSampleAugmentation.h.

References FindKNNIndices(), and SmoteCombine().

SmoteCombine()

SampleType otb::sampleAugmentation::SmoteCombine	(	const SampleType &	s1,
		const SampleType &	s2,
		double	position
	)

Generate the new sample in the line linking s1 and s2

Definition at line 206 of file otbSampleAugmentation.h.

Referenced by Smote().