Struct Conv1d 

pub struct Conv1d {
    pub weight: Tensor,
    pub bias: Option<Tensor>,
    pub stride: usize,
    pub padding: (isize, isize),
}

1D convolution: y = conv1d(x, weight) + bias.

Weight shape: [out_channels, in_channels, kernel], optional bias shape: [out_channels]. stride and padding are stored on the module so Layer::forward stays parameter-free.

Fields

weight: Tensor

bias: Option<Tensor>

stride: usize

padding: (isize, isize)

Implementations

impl Conv1d

pub fn new(weight: Tensor, bias: Option<Tensor>) -> Self

Create a Conv1d from existing weight (and optional bias) tensors.

pub fn with_dims( in_channels: usize, out_channels: usize, kernel: usize, dtype: DType, ) -> Self

Create a Conv1d with deterministic sin() initialization, zero bias.

pub fn with_stride(self, stride: usize) -> Self

pub fn with_padding(self, padding: (isize, isize)) -> Self

Trait Implementations

impl Layer for Conv1d

fn forward(&self, x: &Tensor) -> Result<Tensor>