smile.feature

Class NumericAttributeFeature

java.lang.Object

smile.feature.NumericAttributeFeature

All Implemented Interfaces:

Feature<double[]>

public class NumericAttributeFeature
extends Object
implements Feature<double[]>

Numeric attribute normalization/standardization feature generator. Many machine learning methods such as Neural Networks and SVM with Gaussian kernel also require the features properly scaled/standardized. For example, each variable is scaled into interval [0, 1] or to have mean 0 and standard deviation 1.

Author:

Haifeng Li

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	NumericAttributeFeature.Scaling The types of data scaling.

Constructor Summary

Constructors

Constructor and Description
NumericAttributeFeature(Attribute[] attributes, double[][] data) Constructor.
NumericAttributeFeature(Attribute[] attributes, double lower, double upper, double[][] data) Constructor.
NumericAttributeFeature(Attribute[] attributes, NumericAttributeFeature.Scaling scaling) Constructor.
NumericAttributeFeature(Attribute[] attributes, NumericAttributeFeature.Scaling scaling, double[][] data) Constructor.

Method Summary

Methods

Modifier and Type	Method and Description
Attribute[]	attributes() Returns the variable attributes of generated features.
double	f(double[] object, int id) Generates a feature for given object.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

NumericAttributeFeature

public NumericAttributeFeature(Attribute[] attributes,
                       NumericAttributeFeature.Scaling scaling)

Constructor. Scales numeric attributes into proper range. For logarithm scaling, the attributes must have positive values.

Parameters:

attributes - the variable attributes. Of which, numeric variables will be scaled.

scaling - the way of scaling. The scaling type must be NONE or LOGARITHM because they do not need training data.

NumericAttributeFeature

public NumericAttributeFeature(Attribute[] attributes,
                       NumericAttributeFeature.Scaling scaling,
                       double[][] data)

Constructor. Scales numeric attributes into proper range. For logarithm scaling, the attributes must have positive values. In case of normalization, the min and max values of attributes are used as lower and upper limits. For standardization, variables are scaled to have zero mean and unit variance.

Parameters:

attributes - the variable attributes. Of which, numeric variables will be scaled.

scaling - the way of scaling.

data - the training data to learn scaling parameters.

NumericAttributeFeature

public NumericAttributeFeature(Attribute[] attributes,
                       double lower,
                       double upper,
                       double[][] data)

Constructor. Normalizes numeric attributes with Winsorization: values greater than the specified upper limit are replaced with the upper limit, and those below the lower limit are replace with the lower limit. The specified lower/upper limits are indicate in terms of percentiles of the original distribution.

Parameters:

attributes - the variable attributes. Of which, numeric variables will be normalized.

lower - the lower limit in terms of percentiles of the original distribution (say 5th percentile).

upper - the upper limit in terms of percentiles of the original distribution (say 95th percentile).

data - the training data to learn scaling parameters.

NumericAttributeFeature

public NumericAttributeFeature(Attribute[] attributes,
                       double[][] data)

Constructor. Robustly standardizes numeric attributes by subtracting the median and dividing by the IQR.

Parameters:

attributes - the variable attributes. Of which, numeric variables will be standardized.

data - the training data to learn scaling parameters.

Method Detail

attributes

public Attribute[] attributes()

Description copied from interface: Feature

Returns the variable attributes of generated features. Note that these are NOT the original variable attributes of objects.

Specified by:

attributes in interface Feature<double[]>

Returns:

the variable attributes of generated features

f

public double f(double[] object,
       int id)

Description copied from interface: Feature

Generates a feature for given object.

Specified by:

f in interface Feature<double[]>

Parameters:

object - an object of interest.

id - the index of feature to be generated.

Returns:

a feature value.