DataScaler Class Reference

The DataScaler class computes the scaling coefficients and scales a 2D data matrix with dimensions num_samples by num_features. More...

Inheritance diagram for DataScaler:

Public Types
enum	SCALER_TYPE { NONE, STANDARDIZATION, MEAN_NORMALIZATION, MINMAX_NORMALIZATION }
	Enumeration for supported types of DataScalers.

Public Member Functions
void	scale_samples (const MatrixXd &unscaled_samples, MatrixXd &scaled_samples)
	Apply scaling to a set of unscaled samples. More...
MatrixXd	scale_samples (const MatrixXd &unscaled_samples)
	Apply scaling to a set of unscaled samples. More...
const VectorXd &	get_scaler_features_offsets () const
	Get the vector of offsets. More...
const VectorXd &	get_scaler_features_scale_factors () const
	Get the vector of scaling factors. More...
bool	check_for_zero_scaler_factor (int index)
	Checks an individual scaler feature scale factor for being close to zero; If it is near zero, we can potentially run into a divide-by-zero error if not handled appropriately. More...

Static Public Member Functions
static SCALER_TYPE	scaler_type (const std::string &scaler_name)
	Convert scaler name to enum type. More...

Protected Attributes
bool	hasScaling
	Bool for whether or not the the scaling coefficients have been computed.
RowVectorXd	scaledSample
	Vector for a single scaled sample - (num_features); avoids resize memory allocs.
VectorXd	scalerFeaturesOffsets
	Vector of offsets - (num_features)
VectorXd	scalerFeaturesScaleFactors
	Vector of scaling factors - (num_features)

Private Member Functions
template<class Archive >
void	serialize (Archive &archive, const unsigned int version)
	Serializer for base class data (call from dervied with base_object)

Friends
class	boost::serialization::access
	Allow serializers access to private class data.

Detailed Description

The DataScaler class computes the scaling coefficients and scales a 2D data matrix with dimensions num_samples by num_features.

There are currently 3 scaling options for the DataScaler class:

1. StandardizationScaler - transform each feature to have zero mean and unit variance.
2. NormalizationScaler - normalizies each feature uses the max and min value divided by either the mean value (mean_normalization = true) or min value (mean_normalization = false) Also allows for a norm_factor scaling, required for the direct neural network.
3. NoScaler - scaling coefficients amount to an identity operation

Member Function Documentation

scale_samples() [1/2]

void scale_samples	(	const MatrixXd &	unscaled_samples,
		MatrixXd &	scaled_samples
	)

Apply scaling to a set of unscaled samples.

Parameters

[in]	unscaled_samples	Unscaled matrix of samples
[out]	scaled_samples	Scaled matrix of samples

References DataScaler::check_for_zero_scaler_factor(), DataScaler::scalerFeaturesOffsets, and DataScaler::scalerFeaturesScaleFactors.

Referenced by GaussianProcess::build(), PolynomialRegression::build(), GaussianProcess::covariance(), GaussianProcess::gradient(), PolynomialRegression::gradient(), GaussianProcess::hessian(), PolynomialRegression::hessian(), DataScaler::scale_samples(), GaussianProcess::value(), and PolynomialRegression::value().

scale_samples() [2/2]

MatrixXd scale_samples ( const MatrixXd &  unscaled_samples )

inline

Apply scaling to a set of unscaled samples.

Parameters

[in]

unscaled_samples

Unscaled matrix of samples

Returns

MatrixXd scaled_samples Scaled matrix of samples

References DataScaler::scale_samples().

get_scaler_features_offsets()

const VectorXd& get_scaler_features_offsets ( ) const

inline

Get the vector of offsets.

Returns

Vector of scaler offsets - (num_features)

References DataScaler::scalerFeaturesOffsets.

get_scaler_features_scale_factors()

const VectorXd& get_scaler_features_scale_factors ( ) const

inline

Get the vector of scaling factors.

Returns

Vector of scaling factors - (num_features)

References DataScaler::scalerFeaturesScaleFactors.

Referenced by dakota::surrogates::fd_check_gradient(), and dakota::surrogates::fd_check_hessian().

check_for_zero_scaler_factor()

bool check_for_zero_scaler_factor

(

int

index

)

Checks an individual scaler feature scale factor for being close to zero; If it is near zero, we can potentially run into a divide-by-zero error if not handled appropriately.

Parameters

[in]

index

The scaler feature index to check

Returns

True if the value is near zero

References dakota::near_zero, and DataScaler::scalerFeaturesScaleFactors.

Referenced by DataScaler::scale_samples().

scaler_type()

SCALER_TYPE scaler_type ( const std::string &  scaler_name )

static

Convert scaler name to enum type.

Parameters

[in]

scaler_name

DataScaler name to map

Returns

Corresponding DataScaler enum

References dakota::util::type_name_bimap.

Referenced by GaussianProcess::build(), and PolynomialRegression::build().

---

The documentation for this class was generated from the following files:

• UtilDataScaler.hpp
• UtilDataScaler.cpp