org.deeplearning4j.nn.layers.recurrent

Class BaseRecurrentLayer<LayerConfT extends BaseLayer>

java.lang.Object

org.deeplearning4j.nn.layers.AbstractLayer<LayerConfT>

org.deeplearning4j.nn.layers.BaseLayer<LayerConfT>

org.deeplearning4j.nn.layers.recurrent.BaseRecurrentLayer<LayerConfT>

All Implemented Interfaces:

Serializable, Cloneable, Layer, RecurrentLayer, Model

Direct Known Subclasses:

GravesBidirectionalLSTM, GravesLSTM, LSTM, SimpleRnn

public abstract class BaseRecurrentLayer<LayerConfT extends BaseLayer>
extends BaseLayer<LayerConfT>
implements RecurrentLayer

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from interface org.deeplearning4j.nn.api.Layer

Layer.TrainingMode, Layer.Type

Field Summary

Fields

Modifier and Type	Field and Description
protected Map<String,org.nd4j.linalg.api.ndarray.INDArray>	stateMap stateMap stores the INDArrays needed to do rnnTimeStep() forward pass.
protected Map<String,org.nd4j.linalg.api.ndarray.INDArray>	tBpttStateMap State map for use specifically in truncated BPTT training.

Fields inherited from class org.deeplearning4j.nn.layers.BaseLayer

gradient, gradientsFlattened, gradientViews, optimizer, params, paramsFlattened, score, solver, weightNoiseParams

Fields inherited from class org.deeplearning4j.nn.layers.AbstractLayer

cacheMode, conf, dropoutApplied, dropoutMask, epochCount, index, input, iterationCount, maskArray, maskState, preOutput, trainingListeners

Constructor Summary

Constructors

Constructor and Description
BaseRecurrentLayer(NeuralNetConfiguration conf)
BaseRecurrentLayer(NeuralNetConfiguration conf, org.nd4j.linalg.api.ndarray.INDArray input)

Method Summary

Modifier and Type	Method and Description
void	rnnClearPreviousState() Reset/clear the stateMap for rnnTimeStep() and tBpttStateMap for rnnActivateUsingStoredState()
Map<String,org.nd4j.linalg.api.ndarray.INDArray>	rnnGetPreviousState() Returns a shallow copy of the stateMap
Map<String,org.nd4j.linalg.api.ndarray.INDArray>	rnnGetTBPTTState() Get the RNN truncated backpropagations through time (TBPTT) state for the recurrent layer.
void	rnnSetPreviousState(Map<String,org.nd4j.linalg.api.ndarray.INDArray> stateMap) Set the state map.
void	rnnSetTBPTTState(Map<String,org.nd4j.linalg.api.ndarray.INDArray> state) Set the RNN truncated backpropagations through time (TBPTT) state for the recurrent layer.

Methods inherited from class org.deeplearning4j.nn.layers.BaseLayer

accumulateScore, activate, backpropGradient, calcL1, calcL2, clear, clearNoiseWeightParams, clone, computeGradientAndScore, fit, fit, getGradientsViewArray, getOptimizer, getParam, getParamWithNoise, gradient, hasBias, initParams, layerConf, numParams, params, paramTable, paramTable, preOutput, score, setBackpropGradientsViewArray, setParam, setParams, setParams, setParamsViewArray, setParamTable, setScoreWithZ, toString, transpose, update, update

Methods inherited from class org.deeplearning4j.nn.layers.AbstractLayer

activate, addListeners, applyConstraints, applyDropOutIfNecessary, applyMask, assertInputSet, batchSize, conf, feedForwardMaskArray, getIndex, getInput, getInputMiniBatchSize, getListeners, getMaskArray, gradientAndScore, init, input, layerId, numParams, setCacheMode, setConf, setIndex, setInput, setInputMiniBatchSize, setListeners, setListeners, setMaskArray, type, validateInput

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.deeplearning4j.nn.api.layers.RecurrentLayer

rnnActivateUsingStoredState, rnnTimeStep, tbpttBackpropGradient

Methods inherited from interface org.deeplearning4j.nn.api.Layer

activate, activate, backpropGradient, calcL1, calcL2, clearNoiseWeightParams, clone, feedForwardMaskArray, getEpochCount, getIndex, getInputMiniBatchSize, getIterationCount, getListeners, getMaskArray, isPretrainLayer, setCacheMode, setEpochCount, setIndex, setInput, setInputMiniBatchSize, setIterationCount, setListeners, setListeners, setMaskArray, transpose, type

Methods inherited from interface org.deeplearning4j.nn.api.Model

accumulateScore, addListeners, applyConstraints, batchSize, clear, computeGradientAndScore, conf, fit, fit, getGradientsViewArray, getOptimizer, getParam, gradient, gradientAndScore, init, initParams, input, numParams, numParams, params, paramTable, paramTable, score, setBackpropGradientsViewArray, setConf, setParam, setParams, setParamsViewArray, setParamTable, update, update, validateInput

Field Detail

stateMap

protected Map<String,org.nd4j.linalg.api.ndarray.INDArray> stateMap

stateMap stores the INDArrays needed to do rnnTimeStep() forward pass.

tBpttStateMap

protected Map<String,org.nd4j.linalg.api.ndarray.INDArray> tBpttStateMap

State map for use specifically in truncated BPTT training. Whereas stateMap contains the state from which forward pass is initialized, the tBpttStateMap contains the state at the end of the last truncated bptt

Constructor Detail

BaseRecurrentLayer

public BaseRecurrentLayer(NeuralNetConfiguration conf)

BaseRecurrentLayer

public BaseRecurrentLayer(NeuralNetConfiguration conf,
                          org.nd4j.linalg.api.ndarray.INDArray input)

Method Detail

rnnGetPreviousState

public Map<String,org.nd4j.linalg.api.ndarray.INDArray> rnnGetPreviousState()

Returns a shallow copy of the stateMap

Specified by:

rnnGetPreviousState in interface RecurrentLayer

rnnSetPreviousState

public void rnnSetPreviousState(Map<String,org.nd4j.linalg.api.ndarray.INDArray> stateMap)

Set the state map. Values set using this method will be used in next call to rnnTimeStep()

Specified by:

rnnSetPreviousState in interface RecurrentLayer

rnnClearPreviousState

public void rnnClearPreviousState()

Reset/clear the stateMap for rnnTimeStep() and tBpttStateMap for rnnActivateUsingStoredState()

Specified by:

rnnClearPreviousState in interface RecurrentLayer

rnnGetTBPTTState

public Map<String,org.nd4j.linalg.api.ndarray.INDArray> rnnGetTBPTTState()

Description copied from interface: RecurrentLayer

Get the RNN truncated backpropagations through time (TBPTT) state for the recurrent layer. The TBPTT state is used to store intermediate activations/state between updating parameters when doing TBPTT learning

Specified by:

rnnGetTBPTTState in interface RecurrentLayer

Returns:

State for the RNN layer

rnnSetTBPTTState

public void rnnSetTBPTTState(Map<String,org.nd4j.linalg.api.ndarray.INDArray> state)

Description copied from interface: RecurrentLayer

Set the RNN truncated backpropagations through time (TBPTT) state for the recurrent layer. The TBPTT state is used to store intermediate activations/state between updating parameters when doing TBPTT learning

Specified by:

rnnSetTBPTTState in interface RecurrentLayer

Parameters:

state - TBPTT state to set