GRUSequence#

Versioned name: GRUSequence-5

Category: Sequence processing

Short description: GRUSequence operation represents a series of GRU cells. Each cell is implemented as GRUCell operation.

Detailed description

A single cell in the sequence is implemented in the same way as in GRUCell operation. GRUSequence represents a sequence of GRU cells. The sequence can be connected differently depending on direction attribute that specifies the direction of traversing of input data along sequence dimension or specifies whether it should be a bidirectional sequence. The most of the attributes are in sync with the specification of ONNX GRU operator defined GRUCell

Attributes

  • hidden_size

    • Description: hidden_size specifies hidden state size.

    • Range of values: a positive integer

    • Type: int

    • Required: yes

  • activations

    • Description: activations specifies activation functions for gates, there are two gates, so two activation functions should be specified as a value for this attributes

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

    • Type: a list of strings

    • Default value: sigmoid,tanh

    • Required: no

  • activations_alpha, activations_beta

    • Description: activations_alpha, activations_beta attributes of functions; applicability and meaning of these attributes depends on chosen activation functions

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

    • Type: float[]

    • Default value: None

    • Required: no

  • clip

    • Description: clip specifies bound values [-C, C] for tensor clipping. Clipping is performed 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

  • direction

    • Description: Specify if the RNN is forward, reverse, or bidirectional. If it is one of forward or reverse then num_directions = 1, if it is bidirectional, then num_directions = 2. This num_directions value specifies input/output shape requirements.

    • Range of values: forward, reverse, bidirectional

    • Type: string

    • Required: yes

  • 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 - 3D tensor of type T1 [batch_size, seq_length, input_size], input data. It differs from GRUCell 1st input only by additional axis with size seq_length. Required.

  • 2: initial_hidden_state - 3D tensor of type T1 [batch_size, num_directions, hidden_size], input hidden state data. Required.

  • 3: sequence_lengths - 1D tensor of type T2 [batch_size], specifies real sequence lengths for each batch element. In case of negative values in this input, the operation behavior is undefined. Required.

  • 4: W - 3D tensor of type T1 [num_directions, 3 * hidden_size, input_size], the weights for matrix multiplication, gate order: zrh. Required.

  • 5: R - 3D tensor of type T1 [num_directions, 3 * hidden_size, hidden_size], the recurrence weights for matrix multiplication, gate order: zrh. Required.

  • 6: B - 2D tensor of type T. If linear_before_reset is set to 1, then the shape is [num_directions, 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 [num_directions, 3 * hidden_size], the sum of biases (weights and recurrence weights). Required.

Outputs

  • 1: Y - 4D tensor of type T1 [batch_size, num_directions, seq_len, hidden_size], concatenation of all the intermediate output values of the hidden.

  • 2: Ho - 3D tensor of type T1 [batch_size, num_directions, hidden_size], the last output value of hidden state.

Types

  • T1: any supported floating-point type.

  • T2: any supported integer type.

Example

<layer ... type="GRUSequence" ...>
    <data hidden_size="128"/>
    <input>
        <port id="0">
            <dim>1</dim>
            <dim>4</dim>
            <dim>16</dim>
        </port>
        <port id="1">
            <dim>1</dim>
            <dim>1</dim>
            <dim>128</dim>
        </port>
        <port id="2">
            <dim>1</dim>
        </port>
         <port id="3">
            <dim>1</dim>
            <dim>384</dim>
            <dim>16</dim>
        </port>
         <port id="4">
            <dim>1</dim>
            <dim>384</dim>
            <dim>128</dim>
        </port>
         <port id="5">
            <dim>1</dim>
            <dim>384</dim>
        </port>
    </input>
    <output>
        <port id="6">
            <dim>1</dim>
            <dim>1</dim>
            <dim>4</dim>
            <dim>128</dim>
        </port>
        <port id="7">
            <dim>1</dim>
            <dim>1</dim>
            <dim>128</dim>
        </port>
    </output>
</layer>