org.deeplearning4j.nn.layers.mkldnn

Class MKLDNNBatchNormHelper

java.lang.Object

org.deeplearning4j.nn.layers.mkldnn.MKLDNNBatchNormHelper

All Implemented Interfaces:

LayerHelper, BatchNormalizationHelper

public class MKLDNNBatchNormHelper
extends Object
implements BatchNormalizationHelper

MKL-DNN batch normalization helper implementation

Author:

Alex Black

Field Summary

Fields

Modifier and Type	Field and Description
protected OpContext	context

Constructor Summary

Constructors

Constructor and Description
MKLDNNBatchNormHelper(DataType dataType)

Method Summary

Modifier and Type	Method and Description
Pair<Gradient,INDArray>	backpropGradient(INDArray input, INDArray epsilon, long[] shape, INDArray gamma, INDArray beta, INDArray dGammaView, INDArray dBetaView, double eps, CNN2DFormat format, LayerWorkspaceMgr workspaceMgr)
boolean	checkSupported(double eps, boolean fixedGammaBeta)
INDArray	getMeanCache(DataType dataType)
INDArray	getVarCache(DataType dataType)
Map<String,Long>	helperMemoryUse() Return the currently allocated memory for the helper. (a) Excludes: any shared memory used by multiple helpers/layers (b) Excludes any temporary memory (c) Includes all memory that persists for longer than the helper method This is mainly used for debugging and reporting purposes.
INDArray	preOutput(INDArray x, boolean training, long[] shape, INDArray gamma, INDArray beta, INDArray mean, INDArray var, double decay, double eps, CNN2DFormat format, LayerWorkspaceMgr workspaceMgr)

Methods inherited from class java.lang.Object

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

Field Detail

context

protected OpContext context

Constructor Detail

MKLDNNBatchNormHelper

public MKLDNNBatchNormHelper(DataType dataType)

Method Detail

checkSupported

public boolean checkSupported(double eps,
                              boolean fixedGammaBeta)

Specified by:

checkSupported in interface BatchNormalizationHelper

backpropGradient

public Pair<Gradient,INDArray> backpropGradient(INDArray input,
                                                INDArray epsilon,
                                                long[] shape,
                                                INDArray gamma,
                                                INDArray beta,
                                                INDArray dGammaView,
                                                INDArray dBetaView,
                                                double eps,
                                                CNN2DFormat format,
                                                LayerWorkspaceMgr workspaceMgr)

Specified by:

backpropGradient in interface BatchNormalizationHelper

preOutput

public INDArray preOutput(INDArray x,
                          boolean training,
                          long[] shape,
                          INDArray gamma,
                          INDArray beta,
                          INDArray mean,
                          INDArray var,
                          double decay,
                          double eps,
                          CNN2DFormat format,
                          LayerWorkspaceMgr workspaceMgr)

Specified by:

preOutput in interface BatchNormalizationHelper

getMeanCache

public INDArray getMeanCache(DataType dataType)

Specified by:

getMeanCache in interface BatchNormalizationHelper

getVarCache

public INDArray getVarCache(DataType dataType)

Specified by:

getVarCache in interface BatchNormalizationHelper

helperMemoryUse

public Map<String,Long> helperMemoryUse()

Description copied from interface: LayerHelper

Return the currently allocated memory for the helper.
(a) Excludes: any shared memory used by multiple helpers/layers
(b) Excludes any temporary memory (c) Includes all memory that persists for longer than the helper method
This is mainly used for debugging and reporting purposes. Returns a map:
Key: The name of the type of memory
Value: The amount of memory

Specified by:

helperMemoryUse in interface LayerHelper

Returns:

Map of memory, may be null if none is used.