Struct RNNCell 

pub struct RNNCell<B>
where
    B: Backend,
{
    pub input_size: usize,
    pub hidden_size: usize,
    /* private fields */
}

A single-step vanilla RNN cell.

Computes: h' = tanh(x @ W_ih^T + b_ih + h @ W_hh^T + b_hh)

Shapes

• input x: [batch, input_size]
• hidden h: [batch, hidden_size]
• output h’: [batch, hidden_size]

Fields

input_size: usize

hidden_size: usize

Implementations

impl<B> RNNCell<B>

where B: Backend,

pub fn new( input_size: usize, hidden_size: usize, use_bias: bool, dtype: DType, device: &<B as Backend>::Device, ) -> Result<RNNCell<B>, Error>

Create a new RNNCell with Kaiming uniform initialization.

pub fn forward(&self, x: &Tensor<B>, h: &Tensor<B>) -> Result<Tensor<B>, Error>

Forward: h’ = tanh(x @ W_ih^T + b_ih + h @ W_hh^T + b_hh)

• x: [batch, input_size]
• h: [batch, hidden_size]
• returns h’: [batch, hidden_size]

pub fn parameters(&self) -> Vec<Tensor<B>>

Return all trainable parameters.

pub fn named_parameters(&self) -> Vec<(String, Tensor<B>)>

Return all trainable parameters with names.