org.apache.ignite.ml.preprocessing.encoding.target

Class TargetEncoderPreprocessor<K,V>

java.lang.Object

org.apache.ignite.ml.preprocessing.encoding.EncoderPreprocessor<K,V>

org.apache.ignite.ml.preprocessing.encoding.target.TargetEncoderPreprocessor<K,V>

Type Parameters:

K - Type of a key in upstream data.

V - Type of a value in upstream data.

All Implemented Interfaces:

Serializable, BiFunction<K,V,LabeledVector>, IgniteBiFunction<K,V,LabeledVector>, Preprocessor<K,V>

public class TargetEncoderPreprocessor<K,V>
extends EncoderPreprocessor<K,V>

Preprocessing function that makes Target encoding. The Target Encoder Preprocessor encodes string values (categories) to double values in range [0.0, 1], where the value will be presented as a regularized mean target value of a category. alpha = 1 / (1 + exp(-(categorySize - min_samples_leaf) / beta)) encodedValue = globalTargetMean * (1 - alpha) + categoryTargetMean * alpha if categorySize == 1 then use globalTargetMean min_samples_leaf - minimum samples to take category average into account. beta - smoothing effect to balance categorical average vs prior. Higher value means stronger regularization. ref: https://dl.acm.org/doi/10.1145/507533.507538

This preprocessor can transform multiple columns which indices are handled during training process. These indexes could be defined via .withEncodedFeature(featureIndex) call.

NOTE: it does not add new column but change data in-place.

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected static long	serialVersionUID
protected TargetEncodingMeta[]	targetCounters Filling values.

Fields inherited from class org.apache.ignite.ml.preprocessing.encoding.EncoderPreprocessor

basePreprocessor, encodingValues, handledIndices, KEY_FOR_NULL_VALUES, labelFrequencies

Constructor Summary

Constructors

Constructor and Description
TargetEncoderPreprocessor(TargetEncodingMeta[] targetCounters, Preprocessor<K,V> basePreprocessor, Set<Integer> handledIndices) Constructs a new instance of Frequency Encoder preprocessor.

Method Summary

Modifier and Type	Method and Description
LabeledVector	apply(K k, V v) Applies this preprocessor.

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.apache.ignite.ml.preprocessing.Preprocessor

map

Methods inherited from interface org.apache.ignite.ml.math.functions.IgniteBiFunction

andThen

Methods inherited from interface java.util.function.BiFunction

andThen

Field Detail

serialVersionUID

protected static final long serialVersionUID

See Also:

Constant Field Values

targetCounters

protected final TargetEncodingMeta[] targetCounters

Filling values.

Constructor Detail

TargetEncoderPreprocessor

public TargetEncoderPreprocessor(TargetEncodingMeta[] targetCounters,
                                 Preprocessor<K,V> basePreprocessor,
                                 Set<Integer> handledIndices)

Constructs a new instance of Frequency Encoder preprocessor.

Parameters:

basePreprocessor - Base preprocessor.

handledIndices - Handled indices.

Method Detail

apply

public LabeledVector apply(K k,
                           V v)

Applies this preprocessor.

Parameters:

k - Key.

v - Value.

Returns:

Preprocessed row.