org.deeplearning4j.nn.conf.graph.rnn

Class LastTimeStepVertex

java.lang.Object

org.deeplearning4j.nn.conf.graph.GraphVertex

org.deeplearning4j.nn.conf.graph.rnn.LastTimeStepVertex

All Implemented Interfaces:

Serializable, Cloneable

public class LastTimeStepVertex
extends GraphVertex

LastTimeStepVertex is used in the context of recurrent neural network activations, to go from 3d (time series) activations to 2d activations, by extracting out the last time step of activations for each example.
This can be used for example in sequence to sequence architectures, and potentially for sequence classification. NOTE: Because RNNs may have masking arrays (to allow for examples/time series of different lengths in the same minibatch), it is necessary to provide the same of the network input that has the corresponding mask array. If this input does not have a mask array, the last time step of the input will be used for all examples; otherwise, the time step of the last non-zero entry in the mask array (for each example separately) will be used.

Author:

Alex Black

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
LastTimeStepVertex(String maskArrayInputName)

Method Summary

Modifier and Type	Method and Description
GraphVertex	clone()
boolean	equals(Object o)
MemoryReport	getMemoryReport(InputType... inputTypes) This is a report of the estimated memory consumption for the given vertex
InputType	getOutputType(int layerIndex, InputType... vertexInputs) Determine the type of output for this GraphVertex, given the specified inputs.
int	hashCode()
LastTimeStepVertex	instantiate(ComputationGraph graph, String name, int idx, org.nd4j.linalg.api.ndarray.INDArray paramsView, boolean initializeParams) Create a GraphVertex instance, for the given computation graph, given the configuration instance.
int	maxVertexInputs()
int	minVertexInputs()
int	numParams(boolean backprop)
String	toString()

Methods inherited from class java.lang.Object

finalize, getClass, notify, notifyAll, wait, wait, wait

Constructor Detail

LastTimeStepVertex

public LastTimeStepVertex(String maskArrayInputName)

Parameters:

maskArrayInputName - The name of the input to look at when determining the last time step. Specifically, the mask array of this time series input is used when determining which time step to extract and return.

Method Detail

clone

public GraphVertex clone()

Specified by:

clone in class GraphVertex

equals

public boolean equals(Object o)

Specified by:

equals in class GraphVertex

hashCode

public int hashCode()

Specified by:

hashCode in class GraphVertex

numParams

public int numParams(boolean backprop)

Specified by:

numParams in class GraphVertex

minVertexInputs

public int minVertexInputs()

Specified by:

minVertexInputs in class GraphVertex

Returns:

The Smallest valid number of inputs to this vertex

maxVertexInputs

public int maxVertexInputs()

Specified by:

maxVertexInputs in class GraphVertex

Returns:

The largest valid number of inputs to this vertex

instantiate

public LastTimeStepVertex instantiate(ComputationGraph graph,
                                      String name,
                                      int idx,
                                      org.nd4j.linalg.api.ndarray.INDArray paramsView,
                                      boolean initializeParams)

Description copied from class: GraphVertex

Create a GraphVertex instance, for the given computation graph, given the configuration instance.

Specified by:

instantiate in class GraphVertex

Parameters:

graph - The computation graph that this GraphVertex is to be part of

name - The name of the GraphVertex object

idx - The index of the GraphVertex

paramsView - A view of the full parameters array

initializeParams - If true: initialize the parameters. If false: make no change to the values in the paramsView array

Returns:

The implementation GraphVertex object (i.e., implementation, no the configuration)

getOutputType

public InputType getOutputType(int layerIndex,
                               InputType... vertexInputs)
                        throws InvalidInputTypeException

Description copied from class: GraphVertex

Determine the type of output for this GraphVertex, given the specified inputs. Given that a GraphVertex may do arbitrary processing or modifications of the inputs, the output types can be quite different to the input type(s).
This is generally used to determine when to add preprocessors, as well as the input sizes etc for layers

Specified by:

getOutputType in class GraphVertex

Parameters:

layerIndex - The index of the layer (if appropriate/necessary).

vertexInputs - The inputs to this vertex

Returns:

The type of output for this vertex

Throws:

InvalidInputTypeException - If the input type is invalid for this type of GraphVertex

getMemoryReport

public MemoryReport getMemoryReport(InputType... inputTypes)

Description copied from class: GraphVertex

This is a report of the estimated memory consumption for the given vertex

Specified by:

getMemoryReport in class GraphVertex

Parameters:

inputTypes - Input types to the vertex. Memory consumption is often a function of the input type

Returns:

Memory report for the vertex

toString

public String toString()

Overrides:

toString in class Object