org.deeplearning4j.spark.impl.common.score

Class BaseVaeReconstructionProbWithKeyFunctionAdapter<K>

java.lang.Object

org.deeplearning4j.spark.impl.common.score.BaseVaeScoreWithKeyFunctionAdapter<K>

org.deeplearning4j.spark.impl.common.score.BaseVaeReconstructionProbWithKeyFunctionAdapter<K>

Type Parameters:

K - Type of key, associated with each example. Used to keep track of which score belongs to which example

All Implemented Interfaces:

java.io.Serializable, org.datavec.spark.functions.FlatMapFunctionAdapter<java.util.Iterator<scala.Tuple2<K,org.nd4j.linalg.api.ndarray.INDArray>>,scala.Tuple2<K,java.lang.Double>>

Direct Known Subclasses:

CGVaeReconstructionProbWithKeyFunction

public abstract class BaseVaeReconstructionProbWithKeyFunctionAdapter<K>
extends BaseVaeScoreWithKeyFunctionAdapter<K>

Function to calculate the scores (reconstruction probability or log probability) for a variational autoencoder.
Note that scoring is batched for computational efficiency.

See Also:

Serialized Form

Field Summary

Fields inherited from class org.deeplearning4j.spark.impl.common.score.BaseVaeScoreWithKeyFunctionAdapter

jsonConfig, params

Constructor Summary

Constructors

Constructor and Description
BaseVaeReconstructionProbWithKeyFunctionAdapter(org.apache.spark.broadcast.Broadcast<org.nd4j.linalg.api.ndarray.INDArray> params, org.apache.spark.broadcast.Broadcast<java.lang.String> jsonConfig, boolean useLogProbability, int batchSize, int numSamples)

Method Summary

Modifier and Type	Method and Description
org.nd4j.linalg.api.ndarray.INDArray	computeScore(VariationalAutoencoder vae, org.nd4j.linalg.api.ndarray.INDArray toScore)

Methods inherited from class org.deeplearning4j.spark.impl.common.score.BaseVaeScoreWithKeyFunctionAdapter

call, getVaeLayer

Methods inherited from class java.lang.Object

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

Constructor Detail

BaseVaeReconstructionProbWithKeyFunctionAdapter

public BaseVaeReconstructionProbWithKeyFunctionAdapter(org.apache.spark.broadcast.Broadcast<org.nd4j.linalg.api.ndarray.INDArray> params,
                                                       org.apache.spark.broadcast.Broadcast<java.lang.String> jsonConfig,
                                                       boolean useLogProbability,
                                                       int batchSize,
                                                       int numSamples)

Parameters:

params - MultiLayerNetwork parameters

jsonConfig - MultiLayerConfiguration, as json

useLogProbability - If true: use log probability. False: use raw probability.

batchSize - Batch size to use when scoring

numSamples - Number of samples to use when calling VariationalAutoencoder.reconstructionLogProbability(INDArray, int)

Method Detail

computeScore

public org.nd4j.linalg.api.ndarray.INDArray computeScore(VariationalAutoencoder vae,
                                                         org.nd4j.linalg.api.ndarray.INDArray toScore)

Specified by:

computeScore in class BaseVaeScoreWithKeyFunctionAdapter<K>