Struct LSTM 

pub struct LSTM<B>
where
    B: Backend,
{ /* private fields */ }

A multi-step LSTM that unrolls an LSTMCell over the sequence dimension.

Shapes

• input: [batch, seq_len, input_size]
• output: [batch, seq_len, hidden_size]
• h_n: [batch, hidden_size]
• c_n: [batch, hidden_size]

Implementations

impl<B> LSTM<B>

where B: Backend,

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

Create a new LSTM layer.

pub fn forward( &self, x: &Tensor<B>, hc0: Option<(&Tensor<B>, &Tensor<B>)>, ) -> Result<(Tensor<B>, (Tensor<B>, Tensor<B>)), Error>

Forward pass over the full sequence.

• x: [batch, seq_len, input_size]
• hc0: optional initial (h0, c0), each [batch, hidden_size]. If None, zeros are used.

Returns (output, (h_n, c_n)) where:

• output: [batch, seq_len, hidden_size]
• h_n, c_n: [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.

pub fn cell(&self) -> &LSTMCell<B>

Access the underlying cell.