Package oracle.pgx.api.mllib

Class DeepWalkModelBuilder

java.lang.Object

oracle.pgx.api.mllib.DeepWalkModelBuilder

public class DeepWalkModelBuilder
extends java.lang.Object

Builder for a DeepWalk model. The builder can be used to set the configuration of the model and create the model object.

Constructor Summary

Constructors

Constructor	Description
DeepWalkModelBuilder(PgxSession session, oracle.pgx.api.internal.Core core, java.util.function.Supplier<java.lang.String> keystorePathSupplier, java.util.function.Supplier<char[]> keystorePasswordSupplier)	Please do not use; only meant for being used by PGX itself.

Method Summary

Modifier and Type	Method	Description
DeepWalkModel	build()	Builds the DeepWalk model with the builder configuration
int	getBatchSize()	Gets the batch size for training the model (default: 128)
int	getLayerSize()	Gets the number of dimensions for the output vectors (default: 200)
double	getLearningRate()	Gets the initial learning rate (default: 0.025)
double	getMinLearningRate()	Gets the minimum learning rate (default: 0.0001)
int	getMinWordFrequency()	Gets the minimum word frequency to consider before pruning (default: 1)
int	getNegativeSample()	Gets the number of negative samples (default: 10)
int	getNumEpochs()	Gets the number of epochs to train the model (default: 2)
double	getSampleRate()	Gets the sampling rate (default: 0.00001)
java.lang.Long	getSeed()	Gets the random seed for training the model (default: null)
int	getWalkLength()	Gets the length of the walks (default: 5)
int	getWalksPerVertex()	Gets the number of walks to consider per vertex (default: 4)
int	getWindowSize()	Gets the window size to consider while training the model (default: 5)
boolean	isIgnoreIsolated()	Gets whether the model ignores isolated vertices.
boolean	isShuffle()	Gets whether the model will shuffle the shuffle the random walks before training on them
DeepWalkModelBuilder	setBatchSize(int batchSize)	Sets the batch size for training the model (default: 128)
DeepWalkModelBuilder	setEnableAccelerator(boolean enableAccelerator)	If true, use the accelerator (GPU) if available as the machine learning backend.
DeepWalkModelBuilder	setIgnoreIsolated(boolean ignoreIsolated)	Whether to ignore isolated vertices.
DeepWalkModelBuilder	setLayerSize(int layerSize)	Sets the number of dimensions for the output vectors (default: 200)
DeepWalkModelBuilder	setLearningRate(double lr)	Sets the initial learning rate (default: 0.025)
DeepWalkModelBuilder	setMinLearningRate(double minLr)	Sets the minimum learning rate (default: 0.0001)
DeepWalkModelBuilder	setMinWordFrequency(int minWordFrequency)	Sets the minimum word frequency to consider before pruning (default: 1)
DeepWalkModelBuilder	setNegativeSample(int negativeSample)	Sets the number of negative samples (default: 10)
DeepWalkModelBuilder	setNumEpochs(int numEpochs)	Sets the number of epochs to train the model (default: 2)
DeepWalkModelBuilder	setSampleRate(double sampleRate)	Sets the sample rate (default: 0.0) Used to subsample frequent nodes.
DeepWalkModelBuilder	setSeed(java.lang.Long seed)	Sets the random seed for training the model.
DeepWalkModelBuilder	setShuffle(boolean shuffle)	Sets whether to shuffle the random walks before training on them
DeepWalkModelBuilder	setWalkLength(int walkLength)	Sets the length of the walks (default: 5)
DeepWalkModelBuilder	setWalksPerVertex(int walksPerVertex)	Sets the number of walks to consider per vertex (default: 4)
DeepWalkModelBuilder	setWindowSize(int windowSize)	Sets the window size to consider while training the model (default: 5)

Methods inherited from class java.lang.Object

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

Constructor Detail

DeepWalkModelBuilder

public DeepWalkModelBuilder(PgxSession session,
                            oracle.pgx.api.internal.Core core,
                            java.util.function.Supplier<java.lang.String> keystorePathSupplier,
                            java.util.function.Supplier<char[]> keystorePasswordSupplier)

Please do not use; only meant for being used by PGX itself. Use Analyst.deepWalkModelBuilder() to get a model builder instead.

Parameters:

session - PgxSession to which the model is connected

core - Core to which the model is connected

Method Detail

setMinWordFrequency

public DeepWalkModelBuilder setMinWordFrequency(int minWordFrequency)

Sets the minimum word frequency to consider before pruning (default: 1)

Parameters:

minWordFrequency - minimum word frequency to consider before pruning

Returns:

builder

setShuffle

public DeepWalkModelBuilder setShuffle(boolean shuffle)

Sets whether to shuffle the random walks before training on them

Parameters:

shuffle - whether to shuffle

Returns:

builder

isShuffle

public boolean isShuffle()

Gets whether the model will shuffle the shuffle the random walks before training on them

Returns:

whether the model will shuffle the random walks

getMinWordFrequency

public int getMinWordFrequency()

Gets the minimum word frequency to consider before pruning (default: 1)

Returns:

minimum word frequency to consider before pruning

setSeed

public DeepWalkModelBuilder setSeed(java.lang.Long seed)

Sets the random seed for training the model. If set to null, execution will be non-deterministic.

Note that setting the seed will limit the amount of threads during the learning phase to 1

Parameters:

seed - random seed for training the model

Returns:

builder

getSeed

public java.lang.Long getSeed()

Gets the random seed for training the model (default: null)

Returns:

seed random seed for training the model

setBatchSize

public DeepWalkModelBuilder setBatchSize(int batchSize)

Sets the batch size for training the model (default: 128)

Parameters:

batchSize - batch size for training the model

Returns:

builder

getBatchSize

public int getBatchSize()

Gets the batch size for training the model (default: 128)

Returns:

batchSize batch size for training the model

setNumEpochs

public DeepWalkModelBuilder setNumEpochs(int numEpochs)

Sets the number of epochs to train the model (default: 2)

Parameters:

numEpochs - number of epochs to train the model

Returns:

builder

getNumEpochs

public int getNumEpochs()

Gets the number of epochs to train the model (default: 2)

Returns:

number of epochs to train the model

setLayerSize

public DeepWalkModelBuilder setLayerSize(int layerSize)

Sets the number of dimensions for the output vectors (default: 200)

Parameters:

layerSize - number of dimensions for the output vectors

Returns:

builder

getLayerSize

public int getLayerSize()

Gets the number of dimensions for the output vectors (default: 200)

Returns:

number of dimensions for the output vectors

setLearningRate

public DeepWalkModelBuilder setLearningRate(double lr)

Sets the initial learning rate (default: 0.025)

Parameters:

lr - initial learning rate

Returns:

builder

getLearningRate

public double getLearningRate()

Gets the initial learning rate (default: 0.025)

Returns:

initial learning rate

setMinLearningRate

public DeepWalkModelBuilder setMinLearningRate(double minLr)

Sets the minimum learning rate (default: 0.0001)

Parameters:

minLr - minimum learning rate

Returns:

builder

getMinLearningRate

public double getMinLearningRate()

Gets the minimum learning rate (default: 0.0001)

Returns:

minimum learning rate

setWindowSize

public DeepWalkModelBuilder setWindowSize(int windowSize)

Sets the window size to consider while training the model (default: 5)

Parameters:

windowSize - window size to consider while training the model

Returns:

builder

getWindowSize

public int getWindowSize()

Gets the window size to consider while training the model (default: 5)

Returns:

window size to consider while training the model

setWalkLength

public DeepWalkModelBuilder setWalkLength(int walkLength)

Sets the length of the walks (default: 5)

Parameters:

walkLength - length of the walks

Returns:

builder

getWalkLength

public int getWalkLength()

Gets the length of the walks (default: 5)

Returns:

length of the walks

setWalksPerVertex

public DeepWalkModelBuilder setWalksPerVertex(int walksPerVertex)

Sets the number of walks to consider per vertex (default: 4)

Parameters:

walksPerVertex - number of walks to consider per vertex

Returns:

builder

getWalksPerVertex

public int getWalksPerVertex()

Gets the number of walks to consider per vertex (default: 4)

Returns:

number of walks to consider per vertex

setSampleRate

public DeepWalkModelBuilder setSampleRate(double sampleRate)

Sets the sample rate (default: 0.0) Used to subsample frequent nodes. Nodes are sampled based on their relative frequency computed over all walks. The probability of keeping a node x is max(1.0, P(x) = (sqrt(xFrequency/ sampleRate) + 1) * sampleRate/xFrequency) Smaller values for sample rate mean frequent nodes become less likely to be retained in walks. No subsampling happens if sample rate is 0.0. Values for sample rate should be < 0.4 to have some effect.

Parameters:

sampleRate - the sampling rate

Returns:

builder

getSampleRate

public double getSampleRate()

Gets the sampling rate (default: 0.00001)

Returns:

the sampling rate

setNegativeSample

public DeepWalkModelBuilder setNegativeSample(int negativeSample)

Sets the number of negative samples (default: 10)

Parameters:

negativeSample - the number of negative samples

Returns:

builder

getNegativeSample

public int getNegativeSample()

Gets the number of negative samples (default: 10)

Returns:

the number of negative samples

setIgnoreIsolated

public DeepWalkModelBuilder setIgnoreIsolated(boolean ignoreIsolated)

Whether to ignore isolated vertices. If false, pseudo-walks will consisting of only the node itself will be inserted into the dataset (default: true).

Parameters:

ignoreIsolated - whether to ignore

Returns:

builder

isIgnoreIsolated

public boolean isIgnoreIsolated()

Gets whether the model ignores isolated vertices.

Returns:

whether the model ignores isolated vertices

setEnableAccelerator

public DeepWalkModelBuilder setEnableAccelerator(boolean enableAccelerator)

If true, use the accelerator (GPU) if available as the machine learning backend.

default: {true}

Parameters:

enableAccelerator - whether to use the accelerator if available

Returns:

this

Since:

24.1

build

public DeepWalkModel build()
                    throws java.lang.InterruptedException,
                           java.util.concurrent.ExecutionException

Builds the DeepWalk model with the builder configuration

Returns:

DeepWalk model

Throws:

java.lang.InterruptedException

java.util.concurrent.ExecutionException