org.apache.ignite.ml.dataset.feature.extractor

Class Vectorizer<K,V,C extends Serializable,L>

java.lang.Object

org.apache.ignite.ml.dataset.feature.extractor.Vectorizer<K,V,C,L>

Type Parameters:

K - Type of keys in upstream.

V - Type of values in upstream.

C - Type of "coordinate" - index of feature value in upstream object.

L - Type of label for resulting vectors.

All Implemented Interfaces:

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

Direct Known Subclasses:

ExtractionUtils.DefaultLabelVectorizer, LabeledDummyVectorizer, Vectorizer.VectorizerAdapter

public abstract class Vectorizer<K,V,C extends Serializable,L>
extends Object
implements FeatureLabelExtractor<K,V,L>

Class for extracting labeled vectors from upstream. This is an abstract class providing API for extracting feature and label values by "coordinates" of them from upstream objects. For example BinaryObject can be upstream object and coordinates for them are names of fields with double-values.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	Vectorizer.LabelCoordinate Shotrcuts for coordinates in feature vector.
static class	Vectorizer.VectorizerAdapter<K,V,C extends Serializable,L> Utility class for convenient overridings.

Constructor Summary

Constructors

Constructor and Description
Vectorizer(C... coords) Creates an instance of Vectorizer.

Method Summary

Modifier and Type	Method and Description
protected abstract List<C>	allCoords(K key, V value) Returns list of all coordinate with feature values.
LabeledVector<L>	apply(K key, V value) Extracts labeled vector from upstream object.
protected Vector	createVector(int size) Create an instance of vector.
Vectorizer<K,V,C,L>	exclude(C... coords) Exclude these coordinates from result vector.
LabeledVector<L>	extract(K k, V v) Extract LabeledVector from key and value.
protected abstract Serializable	feature(C coord, K key, V value) Extracts feature value by given coordinate.
protected abstract L	label(C coord, K key, V value) Extract label value by given coordinate.
Vectorizer<K,V,C,L>	labeled(C labelCoord) Sets label coordinate for Vectorizer.
Vectorizer<K,V,C,L>	labeled(Vectorizer.LabelCoordinate labelCoord) Sets label coordinate for Vectorizer.
protected abstract L	zero() Returns default label value for unlabeled data.

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.trainers.FeatureLabelExtractor

extractFeatures, extractLabel

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

Constructor Detail

Vectorizer

public Vectorizer(C... coords)

Creates an instance of Vectorizer.

Parameters:

coords - Coordinates for feature extraction. If array is empty then Vectorizer will extract all fields from upstream object.

Method Detail

apply

public LabeledVector<L> apply(K key,
                              V value)

Extracts labeled vector from upstream object.

Specified by:

apply in interface BiFunction<K,V,LabeledVector>

Parameters:

key - Key.

value - Value.

Returns:

vector.

labeled

public Vectorizer<K,V,C,L> labeled(C labelCoord)

Sets label coordinate for Vectorizer. By default it equals null and zero() will be used as label value.

Parameters:

labelCoord - Label coordinate.

Returns:

this.

labeled

public Vectorizer<K,V,C,L> labeled(Vectorizer.LabelCoordinate labelCoord)

Sets label coordinate for Vectorizer. By default it equals null and zero() will be used as label value.

Parameters:

labelCoord - Label coordinate.

Returns:

this.

exclude

public Vectorizer<K,V,C,L> exclude(C... coords)

Exclude these coordinates from result vector.

Parameters:

coords - Coordinates.

Returns:

this.

extract

public LabeledVector<L> extract(K k,
                                V v)

Extract LabeledVector from key and value.

Specified by:

extract in interface FeatureLabelExtractor<K,V,L>

Parameters:

k - Key.

v - Value.

Returns:

Labeled vector.

feature

protected abstract Serializable feature(C coord,
                                        K key,
                                        V value)

Extracts feature value by given coordinate.

Parameters:

coord - Coordinate.

key - Key.

value - Value.

Returns:

feature value.

label

protected abstract L label(C coord,
                           K key,
                           V value)

Extract label value by given coordinate.

Parameters:

coord - Coordinate.

key - Key.

value - Value.

Returns:

label value.

zero

protected abstract L zero()

Returns default label value for unlabeled data.

Returns:

label value.

allCoords

protected abstract List<C> allCoords(K key,
                                     V value)

Returns list of all coordinate with feature values.

Parameters:

key - Key.

value - Value.

Returns:

all coordinates list.

createVector

protected Vector createVector(int size)

Create an instance of vector.

Parameters:

size - Vector size.

Returns:

vector.