Struct BatchNormLayer 

pub struct BatchNormLayer {
    pub running_mean: Vec<f32>,
    pub running_var: Vec<f32>,
    pub gamma: Vec<f32>,
    pub beta: Vec<f32>,
    pub epsilon: f32,
    pub n_features: usize,
}

Batch normalization layer (inference mode).

Normalizes input features using stored running mean and variance, then applies learned scale (gamma) and shift (beta).

Fields

running_mean: Vec<f32>

Running mean for each feature.

running_var: Vec<f32>

Running variance for each feature.

gamma: Vec<f32>

Learned scale parameter (gamma).

beta: Vec<f32>

Learned shift parameter (beta).

epsilon: f32

Small constant for numerical stability.

n_features: usize

Number of features.

Implementations

impl BatchNormLayer

pub fn new(n_features: usize) -> Self

Create a new batch norm layer with identity transform (gamma=1, beta=0).

pub fn forward(&self, input: &[f32]) -> Vec<f32>

Apply batch normalization in inference mode.

output[i] = gamma[i] * (input[i] - mean[i]) / sqrt(var[i] + eps) + beta[i]

pub fn set_stats(&mut self, mean: &[f32], var: &[f32])

Set the running statistics.

pub fn set_affine(&mut self, gamma: &[f32], beta: &[f32])

Set the affine parameters.

impl BatchNormLayer

pub fn update_running_stats(&mut self, batch: &[Vec<f32>], momentum: f32)

Update running statistics from a mini-batch (training mode).

Uses exponential moving average: running_mean = (1-momentum) * running_mean + momentum * batch_mean

Panics

Panics if batch is empty or if any sample has the wrong feature count.

Trait Implementations

impl Clone for BatchNormLayer

fn clone(&self) -> BatchNormLayer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BatchNormLayer

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

Formats the value using the given formatter.