org.deeplearning4j.nn.conf.graph.rnn

Class ReverseTimeSeriesVertex

java.lang.Object

org.deeplearning4j.nn.conf.graph.GraphVertex

org.deeplearning4j.nn.conf.graph.rnn.ReverseTimeSeriesVertex

All Implemented Interfaces:

Serializable, Cloneable

public class ReverseTimeSeriesVertex
extends GraphVertex

ReverseTimeSeriesVertex is used in recurrent neural networks to revert the order of time series. As a result, the last time step is moved to the beginning of the time series and the first time step is moved to the end. This allows recurrent layers to backward process time series.

Masks: The input might be masked (to allow for varying time series lengths in one minibatch). In this case the present input (mask array = 1) will be reverted in place and the padding (mask array = 0) will be left untouched at the same place. For a time series of length n, this would normally mean, that the first n time steps are reverted and the following padding is left untouched, but more complex masks are supported (e.g. [1, 0, 1, 0, ...].
Note: In order to use mask arrays, the constructor must be called with the name of an network input. The mask of this input is then used in this vertex, too.

Author:

Klaus Broelemann (SCHUFA Holding AG)

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
ReverseTimeSeriesVertex() Creates a new ReverseTimeSeriesVertex that doesn't pay attention to masks
ReverseTimeSeriesVertex(String maskArrayInputName) Creates a new ReverseTimeSeriesVertex that uses the mask array of a given input

Method Summary

Modifier and Type	Method and Description
ReverseTimeSeriesVertex	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()
ReverseTimeSeriesVertex	instantiate(ComputationGraph graph, String name, int idx, INDArray paramsView, boolean initializeParams, org.nd4j.linalg.api.buffer.DataType networkDatatype) Create a GraphVertex instance, for the given computation graph, given the configuration instance.
int	maxVertexInputs()
int	minVertexInputs()
long	numParams(boolean backprop)
String	toString()

Methods inherited from class org.deeplearning4j.nn.conf.graph.GraphVertex

setDataType

Methods inherited from class java.lang.Object

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

Constructor Detail

ReverseTimeSeriesVertex

public ReverseTimeSeriesVertex()

Creates a new ReverseTimeSeriesVertex that doesn't pay attention to masks

ReverseTimeSeriesVertex

public ReverseTimeSeriesVertex(String maskArrayInputName)

Creates a new ReverseTimeSeriesVertex that uses the mask array of a given input

Parameters:

maskArrayInputName - The name of the input that holds the mask.

Method Detail

clone

public ReverseTimeSeriesVertex 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 long 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 ReverseTimeSeriesVertex instantiate(ComputationGraph graph,
                                           String name,
                                           int idx,
                                           INDArray paramsView,
                                           boolean initializeParams,
                                           org.nd4j.linalg.api.buffer.DataType networkDatatype)

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