AttentionWrapperCell

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 apply(input: Tuple[Output, AttentionWrapperState[S, SS, Seq[AS], Seq[ASS]]]): Tuple[Output, AttentionWrapperState[S, SS, Seq[AS], Seq[ASS]]]

   

   Definition Classes

   RNNCell

5. final def asInstanceOf[T0]: T0

   

   Definition Classes

   Any

6. val attentionLayerWeights: Seq[Output]

   

   Attention layer weights to use for projecting the computed attention.

7. val attentions: Seq[Attention[AS, ASS]]

   

   Attention mechanisms to use.

8. val cell: RNNCell[Output, core.Shape, S, SS]

   

   RNN cell being wrapped.

9. val cellInputFn: (Output, Output) ⇒ Output

   

   Function that takes the original cell input tensor and the attention tensor as inputs and returns the mixed cell input to use.

   Function that takes the original cell input tensor and the attention tensor as inputs and returns the mixed cell input to use. Defaults to concatenating the two tensors across their last axis.

10. def clone(): AnyRef

    

    Attributes

    protected[java.lang]

    Definition Classes

    AnyRef

    Annotations

    @throws( ... )

11. final def eq(arg0: AnyRef): Boolean

    

    Definition Classes

    AnyRef

12. def equals(arg0: Any): Boolean

    

    Definition Classes

    AnyRef → Any

13. def finalize(): Unit

    

    Attributes

    protected[java.lang]

    Definition Classes

    AnyRef

    Annotations

    @throws( classOf[java.lang.Throwable] )

14. def forward(input: Tuple[Output, AttentionWrapperState[S, SS, Seq[AS], Seq[ASS]]]): Tuple[Output, AttentionWrapperState[S, SS, Seq[AS], Seq[ASS]]]

    

    Performs a step using this attention-wrapped RNN cell.

    Performs a step using this attention-wrapped RNN cell.

    • Step 1: Mix the inputs and the previous step's attention output via cellInputFn.
    • Step 2: Call the wrapped cell with the mixed input and its previous state.
    • Step 3: Score the cell's output with attentionMechanism.
    • Step 4: Calculate the alignments by passing the score through the normalizer.
    • Step 5: Calculate the context vector as the inner product between the alignments and the attention mechanism's values (memory).
    • Step 6: Calculate the attention output by concatenating the cell output and context through the attention layer (a linear layer with attentionLayerWeights.shape(-1) outputs).

    input

    Input tuple to the attention wrapper cell.

    returns

    Next tuple.

    Definition Classes

    AttentionWrapperCell → RNNCell

15. final def getClass(): Class[_]

    

    Definition Classes

    AnyRef → Any

16. def hashCode(): Int

    

    Definition Classes

    AnyRef → Any

17. def initialState(initialCellState: S, dataType: types.DataType = null): AttentionWrapperState[S, SS, Seq[AS], Seq[ASS]]

    

    Returns an initial state for this attention cell wrapper.

    Returns an initial state for this attention cell wrapper.

    initialCellState

    Initial state for the wrapped cell.

    dataType

    Optional data type which defaults to the data type of the last tensor in initialCellState.

    returns

    Initial state for this attention cell wrapper.

18. final def isInstanceOf[T0]: Boolean

    

    Definition Classes

    Any

19. val name: String

    

    Name prefix used for all new ops.

20. final def ne(arg0: AnyRef): Boolean

    

    Definition Classes

    AnyRef

21. final def notify(): Unit

    

    Definition Classes

    AnyRef

22. final def notifyAll(): Unit

    

    Definition Classes

    AnyRef

23. val outputAttention: Boolean

    

    If true (the default), the output of this cell at each step is the attention value.

    If true (the default), the output of this cell at each step is the attention value. This is the behavior of Luong-style attention mechanisms. If false, the output at each step is the output of cell. This is the behavior of Bhadanau-style attention mechanisms. In both cases, the attention tensor is propagated to the next time step via the state and is used there. This flag only controls whether the attention mechanism is propagated up to the next cell in an RNN stack or to the top RNN output.

24. def outputShape: core.Shape

    

    Definition Classes

    AttentionWrapperCell → RNNCell

25. def stateShape: (SS, core.Shape, core.Shape, Seq[core.Shape], Seq[core.Shape], Seq[ASS])

    

    Definition Classes

    AttentionWrapperCell → RNNCell

26. val storeAlignmentsHistory: Boolean

    

    If true, the alignments history from all steps is stored in the final output state (currently stored as a time major TensorArray on which you must call stack()).

    If true, the alignments history from all steps is stored in the final output state (currently stored as a time major TensorArray on which you must call stack()). Defaults to false.

27. final def synchronized[T0](arg0: ⇒ T0): T0

    

    Definition Classes

    AnyRef

28. def toString(): String

    

    Definition Classes

    AnyRef → Any

29. final def wait(): Unit

    

    Definition Classes

    AnyRef

    Annotations

    @throws( ... )

30. final def wait(arg0: Long, arg1: Int): Unit

    

    Definition Classes

    AnyRef

    Annotations

    @throws( ... )

31. final def wait(arg0: Long): Unit

    

    Definition Classes

    AnyRef

    Annotations

    @throws( ... )

32. def zeroState(batchSize: Output, dataType: types.DataType, shape: (SS, core.Shape, core.Shape, Seq[core.Shape], Seq[core.Shape], Seq[ASS]), name: String = "ZeroState"): AttentionWrapperState[S, SS, Seq[AS], Seq[ASS]]

    

    Definition Classes

    RNNCell