GRUCell

Versioned name: GRUCell-3

Category: Sequence processing

Short description: GRUCell represents a single GRU Cell that computes the output using the formula described in the paper.

Detailed description: GRUCell computes the output Ht for the current time step based on the followint formula:

Formula:
  *  - matrix multiplication
 (.) - Hadamard product(element-wise)
 [,] - concatenation
  f, g - are activation functions.
   zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz)
   rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr)
   ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0
   ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0
   Ht = (1 - zt) (.) ht + zt (.) Ht-1

Attributes

  • hidden_size

    • Description: hidden_size specifies hidden state size.

    • Range of values: a positive integer

    • Type: int

    • Required: yes

  • activations

    • Description: activation functions for gates

    • Range of values: any combination of relu, sigmoid, tanh

    • Type: a list of strings

    • Default value: sigmoid for f, tanh for g

    • Required: no

  • activations_alpha, activations_beta

    • Description: activations_alpha, activations_beta functions attributes

    • Range of values: a list of floating-point numbers

    • Type: float[]

    • Default value: None

    • Required: no

  • clip

    • Description: clip specifies value for tensor clipping to be in [-C, C] before activations

    • Range of values: a positive floating-point number

    • Type: float

    • Default value: infinity that means that the clipping is not applied

    • Required: no

  • linear_before_reset

    • Description: linear_before_reset flag denotes if the layer behaves according to the modification of GRUCell described in the formula in the ONNX documentation.

    • Range of values: true or false

    • Type: boolean

    • Default value: false

    • Required: no

Inputs

  • 1: X - 2D tensor of type T [batch_size, input_size], input data. Required.

  • 2: initial_hidden_state - 2D tensor of type T [batch_size, hidden_size]. Required.

  • 3: W - 2D tensor of type T [3 * hidden_size, input_size], the weights for matrix multiplication, gate order: zrh. Required.

  • 4: R - 2D tensor of type T [3 * hidden_size, hidden_size], the recurrence weights for matrix multiplication, gate order: zrh. Required.

  • 5: B - 1D tensor of type T. If linear_before_reset is set to 1, then the shape is [4 * hidden_size] - the sum of biases for z and r gates (weights and recurrence weights), the biases for h gate are placed separately. Otherwise the shape is [3 * hidden_size], the sum of biases (weights and recurrence weights). Optional.

Outputs

  • 1: Ho - 2D tensor of type T [batch_size, hidden_size], the last output value of hidden state.

Types

  • T: any supported floating-point type.

Example

<layer ... type="GRUCell" ...>
    <data hidden_size="128" linear_before_reset="1"/>
    <input>
        <port id="0">
            <dim>1</dim>
            <dim>16</dim>
        </port>
        <port id="1">
            <dim>1</dim>
            <dim>128</dim>
        </port>
         <port id="2">
            <dim>384</dim>
            <dim>16</dim>
        </port>
         <port id="3">
            <dim>384</dim>
            <dim>128</dim>
        </port>
         <port id="4">
            <dim>768</dim>
        </port>
    </input>
    <output>
        <port id="5">
            <dim>1</dim>
            <dim>128</dim>
        </port>
    </output>
</layer>