org.deeplearning4j.graph.models.deepwalk

Class DeepWalk<V,E>

java.lang.Object

org.deeplearning4j.graph.models.embeddings.GraphVectorsImpl<V,E>

org.deeplearning4j.graph.models.deepwalk.DeepWalk<V,E>

All Implemented Interfaces:

java.io.Serializable, GraphVectors<V,E>

public class DeepWalk<V,E>
extends GraphVectorsImpl<V,E>

Implementation of the DeepWalk graph vectorization model, based on the paper DeepWalk: Online Learning of Social Representations by Perozzi, Al-Rfou & Skiena (2014), http://arxiv.org/abs/1403.6652
Similar to word2vec in nature, DeepWalk is an unsupervised learning algorithm that learns a vector representation of each vertex in a graph. Vector representations are learned using walks (usually random walks) on the vertices in the graph.
Once learned, these vector representations can then be used for purposes such as classification, clustering, similarity search, etc on the graph

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	DeepWalk.Builder<V,E>

Field Summary

Fields

Modifier and Type	Field and Description
static int	STATUS_UPDATE_FREQUENCY

Fields inherited from class org.deeplearning4j.graph.models.embeddings.GraphVectorsImpl

graph, lookupTable

Constructor Summary

Constructors

Constructor and Description
DeepWalk()

Method Summary

Modifier and Type	Method and Description
void	fit(GraphWalkIterator<V> iterator) Fit the DeepWalk model using a single thread using a given GraphWalkIterator.
void	fit(GraphWalkIteratorProvider<V> iteratorProvider) Fit the model, in parallel, using a given GraphWalkIteratorProvider. This object is used to generate multiple GraphWalkIterators, which can then be distributed to each thread to do in parallel Note that fit(IGraph, int) will be more convenient in many cases Note that initialize(IGraph) or initialize(int[]) must be called first.
void	fit(IGraph<V,E> graph, int walkLength) Fit the model, in parallel.
double	getLearningRate()
int	getVectorSize()
int	getWindowSize()
void	initialize(IGraph<V,E> graph) Initialize the DeepWalk model with a given graph.
void	initialize(int[] graphVertexDegrees) Initialize the DeepWalk model with a list of vertex degrees for a graph. Specifically, graphVertexDegrees[i] represents the vertex degree of the ith vertex vertex degrees are used to construct a binary (Huffman) tree, which is in turn used in the hierarchical softmax implementation
GraphVectorLookupTable	lookupTable()
void	setLearningRate(double learningRate)

Methods inherited from class org.deeplearning4j.graph.models.embeddings.GraphVectorsImpl

getGraph, getVertexVector, getVertexVector, numVertices, similarity, similarity, verticesNearest

Methods inherited from class java.lang.Object

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

Field Detail

STATUS_UPDATE_FREQUENCY

public static final int STATUS_UPDATE_FREQUENCY

See Also:

Constant Field Values

Constructor Detail

DeepWalk

public DeepWalk()

Method Detail

getVectorSize

public int getVectorSize()

Specified by:

getVectorSize in interface GraphVectors<V,E>

Overrides:

getVectorSize in class GraphVectorsImpl<V,E>

getWindowSize

public int getWindowSize()

getLearningRate

public double getLearningRate()

setLearningRate

public void setLearningRate(double learningRate)

initialize

public void initialize(IGraph<V,E> graph)

Initialize the DeepWalk model with a given graph.

initialize

public void initialize(int[] graphVertexDegrees)

Initialize the DeepWalk model with a list of vertex degrees for a graph.
Specifically, graphVertexDegrees[i] represents the vertex degree of the ith vertex
vertex degrees are used to construct a binary (Huffman) tree, which is in turn used in the hierarchical softmax implementation

Parameters:

graphVertexDegrees - degrees of each vertex

fit

public void fit(IGraph<V,E> graph,
                int walkLength)

Fit the model, in parallel. This creates a set of GraphWalkIterators, which are then distributed one to each thread

Parameters:

graph - Graph to fit

walkLength - Length of rangom walks to generate

fit

public void fit(GraphWalkIteratorProvider<V> iteratorProvider)

Fit the model, in parallel, using a given GraphWalkIteratorProvider.
This object is used to generate multiple GraphWalkIterators, which can then be distributed to each thread to do in parallel
Note that fit(IGraph, int) will be more convenient in many cases
Note that initialize(IGraph) or initialize(int[]) must be called first.

Parameters:

iteratorProvider - GraphWalkIteratorProvider

See Also:

fit(IGraph, int)

fit

public void fit(GraphWalkIterator<V> iterator)

Fit the DeepWalk model using a single thread using a given GraphWalkIterator. If parallel fitting is required, fit(IGraph, int) or fit(GraphWalkIteratorProvider) should be used.
Note that initialize(IGraph) or initialize(int[]) must be called first.

Parameters:

iterator - iterator for graph walks

lookupTable

public GraphVectorLookupTable lookupTable()