Struct Conv1DLayer 

pub struct Conv1DLayer {
    pub in_channels: usize,
    pub out_channels: usize,
    pub kernel_size: usize,
    pub weights: Vec<Vec<Vec<f64>>>,
    pub biases: Vec<f64>,
    pub activation: ExtActivation,
}

A 1-D convolutional layer operating on a sequence of feature vectors.

Applies a set of out_channels filters, each of length kernel_size spanning in_channels input channels, using causal (left) padding so the output length equals the input length.

Layout:

• weights[o][k][c] = weight for output channel o, kernel position k, input channel c.
• biases[o] = bias for output channel o.

Fields

in_channels: usize

Number of input channels per time step.

out_channels: usize

Number of output channels per time step.

kernel_size: usize

Kernel (filter) length along the time axis.

weights: Vec<Vec<Vec<f64>>>

Filter weights: weights[out_ch][kernel_pos][in_ch].

biases: Vec<f64>

Bias per output channel.

activation: ExtActivation

Activation function applied after convolution.

Implementations

impl Conv1DLayer

pub fn new( in_channels: usize, out_channels: usize, kernel_size: usize, activation: ExtActivation, ) -> Self

Create a new Conv1D layer with zero-initialised weights.

pub fn forward(&self, input: &[Vec<f64>]) -> Vec<Vec<f64>>

Forward pass.

input has shape [seq_len][in_channels]. Returns a tensor of shape [seq_len][out_channels] using causal (left-zero) padding.

pub fn num_params(&self) -> usize

Total number of trainable parameters.

Trait Implementations

impl Clone for Conv1DLayer

fn clone(&self) -> Conv1DLayer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Conv1DLayer

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

Formats the value using the given formatter.