Packages

abstract class AbstractModule[A <: Activity, B <: Activity, T] extends Serializable with InferShape

Module is the basic component of a neural network. It forward activities and backward gradients. Modules can connect to others to construct a complex neural network.

A

Input data type

B

Output data type

T

The numeric type in this module parameters.

Linear Supertypes
InferShape, Serializable, Serializable, AnyRef, Any
Known Subclasses
MaskRCNN, Abs, ActivityRegularization, Add, AddConstant, Attention, BatchNormalization, BiRecurrent, BifurcateSplitTable, Bilinear, BinaryThreshold, BinaryTreeLSTM, Bottle, CAdd, CAddTable, CAveTable, CDivTable, CMaxTable, CMinTable, CMul, CMulTable, CSubTable, Cell, Clamp, Concat, ConcatTable, Container, Contiguous, ConvLSTMPeephole, ConvLSTMPeephole3D, Cosine, CosineDistance, Cropping2D, Cropping3D, CrossProduct, DenseToSparse, DetectionOutputFrcnn, DetectionOutputSSD, DotProduct, Dropout, DynamicContainer, ELU, Echo, Euclidean, Exp, ExpandSize, FlattenTable, GRU, GaussianDropout, GaussianNoise, GaussianSampler, GradientReversal, Graph, HardShrink, HardSigmoid, HardTanh, Identity, Index, InferReshape, Input, JoinTable, L1Penalty, LSTM, LSTMPeephole, LeakyReLU, Linear, LocallyConnected1D, LocallyConnected2D, Log, LogSigmoid, LogSoftMax, LookupTable, LookupTableSparse, MM, MV, MapTable, MaskedSelect, Masking, Max, Maxout, Mean, Min, MixtureTable, Mul, MulConstant, MultiRNNCell, Narrow, NarrowTable, Negative, NegativeEntropyPenalty, Normalize, NormalizeScale, PReLU, Pack, Padding, PairwiseDistance, ParallelTable, Pooler, Power, PriorBox, Proposal, RReLU, ReLU, ReLU6, Recurrent, RecurrentDecoder, RegionProposal, Replicate, Reshape, ResizeBilinear, Reverse, RnnCell, RoiAlign, RoiPooling, SReLU, Scale, Select, SelectTable, SequenceBeamSearch, Sequential, Sigmoid, SoftMax, SoftMin, SoftPlus, SoftShrink, SoftSign, SparseJoinTable, SparseLinear, SpatialAveragePooling, SpatialBatchNormalization, SpatialContrastiveNormalization, SpatialConvolution, SpatialConvolutionMap, SpatialCrossMapLRN, SpatialDilatedConvolution, SpatialDivisiveNormalization, SpatialDropout1D, SpatialDropout2D, SpatialDropout3D, SpatialFullConvolution, SpatialMaxPooling, SpatialSeparableConvolution, SpatialShareConvolution, SpatialSubtractiveNormalization, SpatialWithinChannelLRN, SpatialZeroPadding, SplitTable, Sqrt, Square, Squeeze, StaticGraph, Sum, TableOperation, Tanh, TanhShrink, TemporalConvolution, TemporalMaxPooling, Threshold, Tile, TimeDistributed, Transformer, Transpose, TreeLSTM, Unsqueeze, UpSampling1D, UpSampling2D, UpSampling3D, View, VolumetricAveragePooling, VolumetricConvolution, VolumetricFullConvolution, VolumetricMaxPooling, TensorModule, Activation, AtrousConvolution1D, AtrousConvolution2D, AveragePooling1D, AveragePooling2D, AveragePooling3D, BatchNormalization, Bidirectional, ConvLSTM2D, Convolution1D, Convolution2D, Convolution3D, Cropping1D, Cropping2D, Cropping3D, Deconvolution2D, Dense, Dropout, ELU, Embedding, Flatten, GRU, GaussianDropout, GaussianNoise, GlobalAveragePooling1D, GlobalAveragePooling2D, GlobalAveragePooling3D, GlobalMaxPooling1D, GlobalMaxPooling2D, GlobalMaxPooling3D, GlobalPooling1D, GlobalPooling2D, GlobalPooling3D, Highway, KerasIdentityWrapper, KerasLayer, KerasLayerWrapper, KerasModel, LSTM, LeakyReLU, LocallyConnected1D, LocallyConnected2D, Masking, MaxPooling1D, MaxPooling2D, MaxPooling3D, MaxoutDense, Merge, Permute, Pooling1D, Pooling2D, Pooling3D, Recurrent, RepeatVector, Reshape, SReLU, SeparableConvolution2D, SimpleRNN, SoftMax, SpatialDropout1D, SpatialDropout2D, SpatialDropout3D, ThresholdedReLU, TimeDistributed, UpSampling1D, UpSampling2D, UpSampling3D, ZeroPadding1D, ZeroPadding2D, ZeroPadding3D, AvgPooling, CAddTable, ConcatTable, DnnGraph, Dropout, Identity, Input, JoinTable, LRN, Linear, MaxPooling, MklDnnContainer, MklDnnLayer, Output, RNN, ReLU, ReorderMemory, SelectTable, Sequential, SoftMax, SpatialBatchNormalization, SpatialConvolution, Reshape, All, Any, ApproximateEqual, ArgMax, BatchMatMul, BucketizedCol, Cast, CategoricalColHashBucket, CategoricalColVocaList, Ceil, Compare, CrossCol, CrossEntropy, DepthwiseConv2D, Digamma, Dilation2D, Equal, Erf, Erfc, Exp, Expm1, Floor, FloorDiv, FloorMod, Gather, Greater, GreaterEqual, InTopK, IndicatorCol, Inv, IsFinite, IsInf, IsNan, Kv2Tensor, L2Loss, Less, LessEqual, Lgamma, LogicalAnd, LogicalNot, LogicalOr, Max, Maximum, Minimum, MkString, Mod, ModuleToOperation, NotEqual, OneHot, Operation, Pad, Pow, Prod, RandomUniform, RangeOps, Rank, ResizeBilinearOps, Rint, Round, SegmentSum, Select, SelectTensor, Sign, Slice, SquaredDifference, Substr, Sum, TensorOp, Tile, TopK, TruncateDiv, TruncatedNormal, QuantizedModule, BiasAdd, ControlOps, IdentityControl, Log1p, SqrtGrad, TensorModuleWrapper, Adapter, ConcatV2LoadTF, ExpandDimsLoadTF, MeanLoadTF, PadLoadTF, ProdLoadTF, ReshapeLoadTF, SliceLoadTF, SplitLoadTF, TopKV2LoadTF, TransposeLoadTF, Input, Model, Sequential
Ordering
  1. Alphabetic
  2. By Inheritance
Inherited
  1. AbstractModule
  2. InferShape
  3. Serializable
  4. Serializable
  5. AnyRef
  6. Any
  1. Hide All
  2. Show All
Visibility
  1. Public
  2. All

Instance Constructors

  1. new AbstractModule()(implicit arg0: ClassTag[A], arg1: ClassTag[B], arg2: ClassTag[T], ev: TensorNumeric[T])

Abstract Value Members

  1. abstract def updateGradInput(input: A, gradOutput: B): A

    Computing the gradient of the module with respect to its own input.

    Computing the gradient of the module with respect to its own input. This is returned in gradInput. Also, the gradInput state variable is updated accordingly.

  2. abstract def updateOutput(input: A): B

    Computes the output using the current parameter set of the class and input.

    Computes the output using the current parameter set of the class and input. This function returns the result which is stored in the output field.

Concrete Value Members

  1. final def !=(arg0: Any): Boolean
    Definition Classes
    AnyRef → Any
  2. final def ##(): Int
    Definition Classes
    AnyRef → Any
  3. final def ==(arg0: Any): Boolean
    Definition Classes
    AnyRef → Any
  4. def accGradParameters(input: A, gradOutput: B): Unit

    Computing the gradient of the module with respect to its own parameters.

    Computing the gradient of the module with respect to its own parameters. Many modules do not perform this step as they do not have any parameters. The state variable name for the parameters is module dependent. The module is expected to accumulate the gradients with respect to the parameters in some variable.

  5. def apply(name: String): Option[AbstractModule[Activity, Activity, T]]

    Find a module with given name.

    Find a module with given name. If there is no module with given name, it will return None. If there are multiple modules with the given name, an exception will be thrown.

  6. final def asInstanceOf[T0]: T0
    Definition Classes
    Any
  7. def backward(input: A, gradOutput: B): A

    Performs a back-propagation step through the module, with respect to the given input.

    Performs a back-propagation step through the module, with respect to the given input. In general this method makes the assumption forward(input) has been called before, with the same input. This is necessary for optimization reasons. If you do not respect this rule, backward() will compute incorrect gradients.

    input

    input data

    gradOutput

    gradient of next layer

    returns

    gradient corresponding to input data

  8. var backwardTime: Long
    Attributes
    protected
  9. def clearState(): AbstractModule.this.type

    Clear cached activities to save storage space or network bandwidth.

    Clear cached activities to save storage space or network bandwidth. Note that we use Tensor.set to keep some information like tensor share

    The subclass should override this method if it allocate some extra resource, and call the super.clearState in the override method

  10. final def clone(deepCopy: Boolean): AbstractModule[A, B, T]

    Clone the module, deep or shallow copy

  11. def clone(): AnyRef
    Attributes
    protected[java.lang]
    Definition Classes
    AnyRef
    Annotations
    @native() @throws( ... )
  12. final def cloneModule(): AbstractModule.this.type

    Clone the model

  13. final def eq(arg0: AnyRef): Boolean
    Definition Classes
    AnyRef
  14. def equals(other: Any): Boolean
    Definition Classes
    AbstractModule → AnyRef → Any
  15. final def evaluate(dataSet: LocalDataSet[MiniBatch[T]], vMethods: Array[_ <: ValidationMethod[T]]): Array[(ValidationResult, ValidationMethod[T])]

    use ValidationMethod to evaluate module on the given local dataset

  16. final def evaluate(dataset: RDD[MiniBatch[T]], vMethods: Array[_ <: ValidationMethod[T]]): Array[(ValidationResult, ValidationMethod[T])]

    use ValidationMethod to evaluate module on the given rdd dataset

  17. final def evaluate(dataset: RDD[Sample[T]], vMethods: Array[_ <: ValidationMethod[T]], batchSize: Option[Int] = None): Array[(ValidationResult, ValidationMethod[T])]

    use ValidationMethod to evaluate module on the given rdd dataset

    use ValidationMethod to evaluate module on the given rdd dataset

    dataset

    dataset for test

    vMethods

    validation methods

    batchSize

    total batchsize of all partitions, optional param and default 4 * partitionNum of dataset

  18. def evaluate(): AbstractModule.this.type

    Set the module to evaluate mode

  19. final def evaluateImage(imageFrame: ImageFrame, vMethods: Array[_ <: ValidationMethod[T]], batchSize: Option[Int] = None): Array[(ValidationResult, ValidationMethod[T])]

    use ValidationMethod to evaluate module on the given ImageFrame

    use ValidationMethod to evaluate module on the given ImageFrame

    imageFrame

    ImageFrame for valudation

    vMethods

    validation methods

    batchSize

    total batch size of all partitions

  20. def finalize(): Unit
    Attributes
    protected[java.lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( classOf[java.lang.Throwable] )
  21. final def forward(input: A): B

    Takes an input object, and computes the corresponding output of the module.

    Takes an input object, and computes the corresponding output of the module. After a forward, the output state variable should have been updated to the new value.

    input

    input data

    returns

    output data

  22. var forwardTime: Long
    Attributes
    protected
  23. def freeze(names: String*): AbstractModule.this.type

    freeze the module, i.e.

    freeze the module, i.e. their parameters(weight/bias, if exists) are not changed in training process if names is not empty, set an array of layers that match the given names to be "freezed",

    names

    an array of layer names

    returns

    current graph model

  24. final def getClass(): Class[_]
    Definition Classes
    AnyRef → Any
    Annotations
    @native()
  25. def getExtraParameter(): Array[Tensor[T]]

    Get extra parameter in this module.

    Get extra parameter in this module. Extra parameter means the trainable parameters beside weight and bias. Such as runningMean and runningVar in BatchNormalization.

    The subclass should override this method if it has some parameters besides weight and bias.

    returns

    an array of tensor

  26. final def getInputShape(): Shape

    Return the inputShape for the current Layer and the first dim is batch.

    Return the inputShape for the current Layer and the first dim is batch.

    Definition Classes
    InferShape
  27. final def getName(): String

    Get the module name, default name is className@namePostfix

  28. final def getNumericType(): TensorDataType

    Get numeric type of module parameters

  29. final def getOutputShape(): Shape

    Return the outputShape for the current Layer and the first dim is batch.

    Return the outputShape for the current Layer and the first dim is batch.

    Definition Classes
    InferShape
  30. def getParametersTable(): Table

    This function returns a table contains ModuleName, the parameter names and parameter value in this module.

    This function returns a table contains ModuleName, the parameter names and parameter value in this module.

    The result table is a structure of Table(ModuleName -> Table(ParameterName -> ParameterValue)), and the type is Table[String, Table[String, Tensor[T]]].

    For example, get the weight of a module named conv1: table[Table]("conv1")[Tensor[T]]("weight").

    The names of the parameters follow such convention:

    1. If there's one parameter, the parameter is named as "weight", the gradient is named as "gradWeight"

    2. If there're two parameters, the first parameter is named as "weight", the first gradient is named as "gradWeight"; the second parameter is named as "bias", the seconcd gradient is named as "gradBias"

    3. If there're more parameters, the weight is named as "weight" with a seq number as suffix, the gradient is named as "gradient" with a seq number as suffix

    Custom modules should override this function the default impl if the convention doesn't meet the requirement.

    returns

    Table

  31. final def getPrintName(): String
    Attributes
    protected
  32. final def getScaleB(): Double

    Get the scale of gradientBias

  33. final def getScaleW(): Double

    Get the scale of gradientWeight

  34. def getTimes(): Array[(AbstractModule[_ <: Activity, _ <: Activity, T], Long, Long)]

    Get the forward/backward cost time for the module or its submodules

  35. final def getTimesGroupByModuleType(): Array[(String, Long, Long)]

    Get the forward/backward cost time for the module or its submodules and group by module type.

    Get the forward/backward cost time for the module or its submodules and group by module type.

    returns

    (module type name, forward time, backward time)

  36. final def getWeightsBias(): Array[Tensor[T]]

    Get weight and bias for the module

    Get weight and bias for the module

    returns

    array of weights and bias

  37. var gradInput: A

    The cached gradient of activities.

    The cached gradient of activities. So we don't compute it again when need it

  38. final def hasName: Boolean

    Whether user set a name to the module before

  39. def hashCode(): Int
    Definition Classes
    AbstractModule → AnyRef → Any
  40. def inputs(first: (ModuleNode[T], Int), nodesWithIndex: (ModuleNode[T], Int)*): ModuleNode[T]

    Build graph: some other modules point to current module

    Build graph: some other modules point to current module

    first

    distinguish from another inputs when input parameter list is empty

    nodesWithIndex

    upstream module nodes and the output tensor index. The start index is 1.

    returns

    node containing current module

  41. def inputs(nodes: Array[ModuleNode[T]]): ModuleNode[T]

    Build graph: some other modules point to current module

    Build graph: some other modules point to current module

    nodes

    upstream module nodes in an array

    returns

    node containing current module

  42. def inputs(nodes: ModuleNode[T]*): ModuleNode[T]

    Build graph: some other modules point to current module

    Build graph: some other modules point to current module

    nodes

    upstream module nodes

    returns

    node containing current module

  43. var inputsFormats: Seq[Int]
    Attributes
    protected
  44. final def isInstanceOf[T0]: Boolean
    Definition Classes
    Any
  45. final def isTraining(): Boolean

    Check if the model is in training mode

  46. var line: String
    Attributes
    protected
  47. final def loadModelWeights(srcModel: Module[Float], matchAll: Boolean = true): AbstractModule.this.type

    copy weights from another model, mapping by layer name

    copy weights from another model, mapping by layer name

    srcModel

    model to copy from

    matchAll

    whether to match all layers' weights and bias,

    returns

    current module

  48. final def loadWeights(weightPath: String, matchAll: Boolean = true): AbstractModule.this.type

    load pretrained weights and bias to current module

    load pretrained weights and bias to current module

    weightPath

    file to store weights and bias

    matchAll

    whether to match all layers' weights and bias, if not, only load existing pretrained weights and bias

    returns

    current module

  49. final def ne(arg0: AnyRef): Boolean
    Definition Classes
    AnyRef
  50. final def notify(): Unit
    Definition Classes
    AnyRef
    Annotations
    @native()
  51. final def notifyAll(): Unit
    Definition Classes
    AnyRef
    Annotations
    @native()
  52. var output: B

    The cached output.

    The cached output. So we don't compute it again when need it

  53. var outputsFormats: Seq[Int]
    Attributes
    protected
  54. def parameters(): (Array[Tensor[T]], Array[Tensor[T]])

    This function returns two arrays.

    This function returns two arrays. One for the weights and the other the gradients Custom modules should override this function if they have parameters

    returns

    (Array of weights, Array of grad)

  55. final def predict(dataset: RDD[Sample[T]], batchSize: Int = -1, shareBuffer: Boolean = false): RDD[Activity]

    module predict, return the probability distribution

    module predict, return the probability distribution

    dataset

    dataset for prediction

    batchSize

    total batchSize for all partitions. if -1, default is 4 * partitionNumber of datatset

    shareBuffer

    whether to share same memory for each batch predict results

  56. final def predictClass(dataset: RDD[Sample[T]], batchSize: Int = -1): RDD[Int]

    module predict, return the predict label

    module predict, return the predict label

    dataset

    dataset for prediction

    batchSize

    total batchSize for all partitions. if -1, default is 4 * partitionNumber of dataset

  57. final def predictImage(imageFrame: ImageFrame, outputLayer: String = null, shareBuffer: Boolean = false, batchPerPartition: Int = 4, predictKey: String = ImageFeature.predict, featurePaddingParam: Option[PaddingParam[T]] = None): ImageFrame

    model predict images, return imageFrame with predicted tensor, if you want to call predictImage multiple times, it is recommended to use Predictor for DistributedImageFrame or LocalPredictor for LocalImageFrame

    model predict images, return imageFrame with predicted tensor, if you want to call predictImage multiple times, it is recommended to use Predictor for DistributedImageFrame or LocalPredictor for LocalImageFrame

    imageFrame

    imageFrame that contains images

    outputLayer

    if outputLayer is not null, the output of layer that matches outputLayer will be used as predicted output

    shareBuffer

    whether to share same memory for each batch predict results

    batchPerPartition

    batch size per partition, default is 4

    predictKey

    key to store predicted result

    featurePaddingParam

    featurePaddingParam if the inputs have variant size

  58. def processInputs(first: (ModuleNode[T], Int), nodesWithIndex: (ModuleNode[T], Int)*): ModuleNode[T]
    Attributes
    protected
  59. def processInputs(nodes: Seq[ModuleNode[T]]): ModuleNode[T]
    Attributes
    protected
  60. final def quantize(): Module[T]

    Quantize this module, which reduces the precision of the parameter.

    Quantize this module, which reduces the precision of the parameter. Get a higher speed with a little accuracy cost.

  61. def release(): Unit

    if the model contains native resources such as aligned memory, we should release it by manual.

    if the model contains native resources such as aligned memory, we should release it by manual. JVM GC can't release them reliably.

  62. def reset(): Unit

    Reset module parameters, which is re-initialize the parameter with given initMethod

  63. def resetTimes(): Unit

    Reset the forward/backward record time for the module or its submodules

  64. final def saveCaffe(prototxtPath: String, modelPath: String, useV2: Boolean = true, overwrite: Boolean = false): AbstractModule.this.type

    Save this module to path in caffe readable format

  65. final def saveDefinition(path: String, overWrite: Boolean = false): AbstractModule.this.type

    Save this module definition to path.

    Save this module definition to path.

    path

    path to save module, local file system, HDFS and Amazon S3 is supported. HDFS path should be like "hdfs://[host]:[port]/xxx" Amazon S3 path should be like "s3a://bucket/xxx"

    overWrite

    if overwrite

    returns

    self

  66. final def saveModule(path: String, weightPath: String = null, overWrite: Boolean = false): AbstractModule.this.type

    Save this module to path with protobuf format

    Save this module to path with protobuf format

    path

    path to save module, local file system, HDFS and Amazon S3 is supported. HDFS path should be like "hdfs://[host]:[port]/xxx" Amazon S3 path should be like "s3a://bucket/xxx"

    weightPath

    where to store weight

    overWrite

    if overwrite

    returns

    self

  67. final def saveTF(inputs: Seq[(String, Seq[Int])], path: String, byteOrder: ByteOrder = ByteOrder.LITTLE_ENDIAN, dataFormat: TensorflowDataFormat = TensorflowDataFormat.NHWC): AbstractModule.this.type

    Save this module to path in tensorflow readable format

  68. final def saveTorch(path: String, overWrite: Boolean = false): AbstractModule.this.type

    Save this module to path in torch7 readable format

  69. final def saveWeights(path: String, overWrite: Boolean): Unit

    save weights and bias to file

    save weights and bias to file

    path

    file to save

    overWrite

    whether to overwrite or not

  70. var scaleB: Double
    Attributes
    protected
  71. var scaleW: Double

    The scale of gradient weight and gradient bias before gradParameters being accumulated.

    The scale of gradient weight and gradient bias before gradParameters being accumulated.

    Attributes
    protected
  72. def setExtraParameter(extraParam: Array[Tensor[T]]): AbstractModule.this.type

    Set extra parameter to this module.

    Set extra parameter to this module. Extra parameter means the trainable parameters beside weight and bias. Such as runningMean and runningVar in BatchNormalization.

    returns

    this

  73. def setInputFormats(formats: Seq[Int]): AbstractModule.this.type

    set input formats for graph

  74. final def setLine(line: String): AbstractModule.this.type

    Set the line separator when print the module

  75. final def setName(name: String): AbstractModule.this.type

    Set the module name

  76. def setOutputFormats(formats: Seq[Int]): AbstractModule.this.type

    set output formats for graph

  77. def setScaleB(b: Double): AbstractModule.this.type

    Set the scale of gradientBias

    Set the scale of gradientBias

    b

    the value of the scale of gradientBias

    returns

    this

  78. def setScaleW(w: Double): AbstractModule.this.type

    Set the scale of gradientWeight

    Set the scale of gradientWeight

    w

    the value of the scale of gradientWeight

    returns

    this

  79. final def setWeightsBias(newWeights: Array[Tensor[T]]): AbstractModule.this.type

    Set weight and bias for the module

    Set weight and bias for the module

    newWeights

    array of weights and bias

  80. final def synchronized[T0](arg0: ⇒ T0): T0
    Definition Classes
    AnyRef
  81. def toGraph(startNodes: ModuleNode[T]*): Graph[T]

    Generate graph module with start nodes

  82. def toString(): String
    Definition Classes
    AbstractModule → AnyRef → Any
  83. var train: Boolean

    Module status.

    Module status. It is useful for modules like dropout/batch normalization

    Attributes
    protected
  84. def training(): AbstractModule.this.type

    Set the module to training mode

  85. def unFreeze(names: String*): AbstractModule.this.type

    "unfreeze" module, i.e.

    "unfreeze" module, i.e. make the module parameters(weight/bias, if exists) to be trained(updated) in training process if names is not empty, unfreeze layers that match given names

    names

    array of module names to unFreeze

  86. final def wait(): Unit
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )
  87. final def wait(arg0: Long, arg1: Int): Unit
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )
  88. final def wait(arg0: Long): Unit
    Definition Classes
    AnyRef
    Annotations
    @native() @throws( ... )
  89. def zeroGradParameters(): Unit

    If the module has parameters, this will zero the accumulation of the gradients with respect to these parameters.

    If the module has parameters, this will zero the accumulation of the gradients with respect to these parameters. Otherwise, it does nothing.

Deprecated Value Members

  1. def save(path: String, overWrite: Boolean = false): AbstractModule.this.type

    Save this module to path.

    Save this module to path.

    path

    path to save module, local file system, HDFS and Amazon S3 is supported. HDFS path should be like "hdfs://[host]:[port]/xxx" Amazon S3 path should be like "s3a://bucket/xxx"

    overWrite

    if overwrite

    returns

    self

    Annotations
    @deprecated
    Deprecated

    (Since version 0.3.0) please use recommended saveModule(path, overWrite)

Inherited from InferShape

Inherited from Serializable

Inherited from Serializable

Inherited from AnyRef

Inherited from Any

Ungrouped