Struct GruConfig 

pub struct GruConfig {
    pub d_input: usize,
    pub d_hidden: usize,
    pub bias: bool,
    pub reset_after: bool,
    pub initializer: Initializer,
    pub gate_activation: ActivationConfig,
    pub hidden_activation: ActivationConfig,
    pub clip: Option<f64>,
}

Configuration to create a gru module using the init function.

Fields

d_input: usize

The size of the input features.

d_hidden: usize

The size of the hidden state.

bias: bool

If a bias should be applied during the Gru transformation.

reset_after: bool

If reset gate should be applied after weight multiplication.

This configuration option controls how the reset gate is applied to the hidden state.

• true - (Default) Match the initial arXiv version of the paper Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (v1) and apply the reset gate after multiplication by the weights. This matches the behavior of PyTorch GRU.
• false - Match the most recent revision of Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (v3) and apply the reset gate before the weight multiplication.

The differing implementations can give slightly different numerical results and have different efficiencies. For more motivation for why the true can be more efficient see Optimizing RNNs with Differentiable Graphs.

To set this field to false use with_reset_after.

initializer: Initializer

Gru initializer

gate_activation: ActivationConfig

Activation function for the update and reset gates. Default is Sigmoid, which is standard for GRU gates.

hidden_activation: ActivationConfig

Activation function for the new/candidate gate. Default is Tanh, which is standard for GRU.

clip: Option<f64>

Optional hidden state clip threshold. If provided, hidden state values are clipped to the range [-clip, +clip] after each timestep. This can help prevent exploding values during inference.

Implementations

impl GruConfig

pub fn new(d_input: usize, d_hidden: usize, bias: bool) -> GruConfig

Create a new instance of the config.

Arguments

Required Arguments

d_input

The size of the input features.

d_hidden

The size of the hidden state.

bias

If a bias should be applied during the Gru transformation.

Optional Arguments

clip

Optional hidden state clip threshold. If provided, hidden state values are clipped to the range [-clip, +clip] after each timestep. This can help prevent exploding values during inference.

• Defaults to None

Default Arguments

reset_after

If reset gate should be applied after weight multiplication.

This configuration option controls how the reset gate is applied to the hidden state.

• true - (Default) Match the initial arXiv version of the paper Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (v1) and apply the reset gate after multiplication by the weights. This matches the behavior of PyTorch GRU.
• false - Match the most recent revision of Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (v3) and apply the reset gate before the weight multiplication.

The differing implementations can give slightly different numerical results and have different efficiencies. For more motivation for why the true can be more efficient see Optimizing RNNs with Differentiable Graphs.

To set this field to false use with_reset_after.

• Defaults to "true"

initializer

Gru initializer

• Defaults to "Initializer::XavierNormal{gain:1.0}"

gate_activation

Activation function for the update and reset gates. Default is Sigmoid, which is standard for GRU gates.

• Defaults to "ActivationConfig::Sigmoid"

hidden_activation

Activation function for the new/candidate gate. Default is Tanh, which is standard for GRU.

• Defaults to "ActivationConfig::Tanh"

impl GruConfig

pub fn with_reset_after(self, reset_after: bool) -> GruConfig

Sets the value for the field reset_after.

If reset gate should be applied after weight multiplication.

This configuration option controls how the reset gate is applied to the hidden state.

• true - (Default) Match the initial arXiv version of the paper Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (v1) and apply the reset gate after multiplication by the weights. This matches the behavior of PyTorch GRU.
• false - Match the most recent revision of Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (v3) and apply the reset gate before the weight multiplication.

The differing implementations can give slightly different numerical results and have different efficiencies. For more motivation for why the true can be more efficient see Optimizing RNNs with Differentiable Graphs.

To set this field to false use with_reset_after.

• Defaults to "true"

pub fn with_initializer(self, initializer: Initializer) -> GruConfig

Sets the value for the field initializer.

Gru initializer

• Defaults to "Initializer::XavierNormal{gain:1.0}"

pub fn with_gate_activation( self, gate_activation: ActivationConfig, ) -> GruConfig

Sets the value for the field gate_activation.

Activation function for the update and reset gates. Default is Sigmoid, which is standard for GRU gates.

• Defaults to "ActivationConfig::Sigmoid"

pub fn with_hidden_activation( self, hidden_activation: ActivationConfig, ) -> GruConfig

Sets the value for the field hidden_activation.

Activation function for the new/candidate gate. Default is Tanh, which is standard for GRU.

• Defaults to "ActivationConfig::Tanh"

pub fn with_clip(self, clip: Option<f64>) -> GruConfig

Sets the value for the field clip.

Optional hidden state clip threshold. If provided, hidden state values are clipped to the range [-clip, +clip] after each timestep. This can help prevent exploding values during inference.

• Defaults to None

impl GruConfig

pub fn init<B>(&self, device: &<B as BackendTypes>::Device) -> Gru<B>

where B: Backend,

Initialize a new gru module.

Trait Implementations

impl Clone for GruConfig

fn clone(&self) -> GruConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Config for GruConfig

fn save<P>(&self, file: P) -> Result<(), Error>

where P: AsRef<Path>,

Saves the configuration to a file.

fn load<P>(file: P) -> Result<Self, ConfigError>

where P: AsRef<Path>,

Loads the configuration from a file.

fn load_binary(data: &[u8]) -> Result<Self, ConfigError>

Loads the configuration from a binary buffer.

impl Debug for GruConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for GruConfig

fn deserialize<D>( deserializer: D, ) -> Result<GruConfig, <D as Deserializer<'de>>::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for GruConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Serialize for GruConfig

fn serialize<S>( &self, serializer: S, ) -> Result<<S as Serializer>::Ok, <S as Serializer>::Error>

where S: Serializer,

Serialize this value into the given Serde serializer.