org.deeplearning4j.nn.graph.vertex.impl

Class ShiftVertex

java.lang.Object

org.deeplearning4j.nn.graph.vertex.BaseGraphVertex

org.deeplearning4j.nn.graph.vertex.impl.ShiftVertex

All Implemented Interfaces:

Serializable, GraphVertex

public class ShiftVertex
extends BaseGraphVertex

A ShiftVertex is used to shift the activations of a single layer
One could use it to add a bias or as part of some other calculation. For example, Highway Layers need them in two places. One, it's often useful to have the gate weights have a large negative bias. (Of course for this, we could just initialize the biases that way.) But, _also_ it needs to do this: (1-sigmoid(weight * input + bias)) (*) input + sigmoid(weight * input + bias) (*) activation(w2 * input + bias) ((*) is hadamard product) So, here, we could have 1. a DenseLayer that does the sigmoid 2. a ScaleVertex(-1) and 3. a ShiftVertex(1) to accomplish that.

Author:

Binesh Bannerjee ([email protected], @bnsh on gitter)

See Also:

Serialized Form

Field Summary

Fields inherited from class org.deeplearning4j.nn.graph.vertex.BaseGraphVertex

epsilon, graph, inputs, inputVertices, outputVertex, outputVertices, vertexIndex, vertexName

Constructor Summary

Constructors

Constructor and Description
ShiftVertex(ComputationGraph graph, String name, int vertexIndex, double shiftFactor)
ShiftVertex(ComputationGraph graph, String name, int vertexIndex, VertexIndices[] inputVertices, VertexIndices[] outputVertices, double shiftFactor)

Method Summary

Modifier and Type	Method and Description
org.nd4j.linalg.primitives.Pair<Gradient,org.nd4j.linalg.api.ndarray.INDArray[]>	doBackward(boolean tbptt, LayerWorkspaceMgr workspaceMgr) Do backward pass
org.nd4j.linalg.api.ndarray.INDArray	doForward(boolean training, LayerWorkspaceMgr workspaceMgr) Do forward pass using the stored inputs
org.nd4j.linalg.primitives.Pair<org.nd4j.linalg.api.ndarray.INDArray,MaskState>	feedForwardMaskArrays(org.nd4j.linalg.api.ndarray.INDArray[] maskArrays, MaskState currentMaskState, int minibatchSize)
Layer	getLayer() Get the Layer (if any).
boolean	hasLayer() Whether the GraphVertex contains a Layer object or not
void	setBackpropGradientsViewArray(org.nd4j.linalg.api.ndarray.INDArray backpropGradientsViewArray) See Model.setBackpropGradientsViewArray(INDArray)
String	toString()

Methods inherited from class org.deeplearning4j.nn.graph.vertex.BaseGraphVertex

canDoBackward, canDoForward, clear, clearVertex, getEpsilon, getInputVertices, getNumInputArrays, getNumOutputConnections, getOutputVertices, getVertexIndex, getVertexName, isInputVertex, setEpsilon, setInput, setInputVertices, setLayerAsFrozen, setOutputVertices

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.deeplearning4j.nn.graph.vertex.GraphVertex

getInputs, isOutputVertex, setInputs, setOutputVertex

Constructor Detail

ShiftVertex

public ShiftVertex(ComputationGraph graph,
                   String name,
                   int vertexIndex,
                   double shiftFactor)

ShiftVertex

public ShiftVertex(ComputationGraph graph,
                   String name,
                   int vertexIndex,
                   VertexIndices[] inputVertices,
                   VertexIndices[] outputVertices,
                   double shiftFactor)

Method Detail

hasLayer

public boolean hasLayer()

Description copied from interface: GraphVertex

Whether the GraphVertex contains a Layer object or not

getLayer

public Layer getLayer()

Description copied from interface: GraphVertex

Get the Layer (if any). Returns null if GraphVertex.hasLayer() == false

doForward

public org.nd4j.linalg.api.ndarray.INDArray doForward(boolean training,
                                                      LayerWorkspaceMgr workspaceMgr)

Description copied from interface: GraphVertex

Do forward pass using the stored inputs

Parameters:

training - if true: forward pass at training time. If false: forward pass at test time

Returns:

The output (for example, activations) of the GraphVertex

doBackward

public org.nd4j.linalg.primitives.Pair<Gradient,org.nd4j.linalg.api.ndarray.INDArray[]> doBackward(boolean tbptt,
                                                                                                   LayerWorkspaceMgr workspaceMgr)

Description copied from interface: GraphVertex

Do backward pass

Parameters:

tbptt - If true: do backprop using truncated BPTT

Returns:

The gradients (may be null), and the errors/epsilons for all inputs to this GraphVertex

setBackpropGradientsViewArray

public void setBackpropGradientsViewArray(org.nd4j.linalg.api.ndarray.INDArray backpropGradientsViewArray)

Description copied from interface: GraphVertex

See Model.setBackpropGradientsViewArray(INDArray)

toString

public String toString()

Specified by:

toString in class BaseGraphVertex

feedForwardMaskArrays

public org.nd4j.linalg.primitives.Pair<org.nd4j.linalg.api.ndarray.INDArray,MaskState> feedForwardMaskArrays(org.nd4j.linalg.api.ndarray.INDArray[] maskArrays,
                                                                                                             MaskState currentMaskState,
                                                                                                             int minibatchSize)