Struct Linear 

pub struct Linear { /* private fields */ }

Fully connected layer: y = xW^T + b

Applies a linear transformation to the incoming data. Weight initialization follows Xavier/Glorot (Glorot & Bengio, 2010).

Shape

• Input: (*, in_features) where * means any number of batch dimensions
• Output: (*, out_features)

Example

use aprender::nn::{Module, Linear};
use aprender::autograd::Tensor;

let layer = Linear::new(20, 30);  // 20 inputs, 30 outputs
let x = Tensor::randn(&[128, 20]);  // batch of 128
let output = layer.forward(&x);     // [128, 30]

assert_eq!(output.shape(), &[128, 30]);

Implementations

impl Linear

pub fn new(in_features: usize, out_features: usize) -> Self

Create a new Linear layer with Xavier initialization.

Arguments

• in_features - Number of input features
• out_features - Number of output features

Example

let layer = Linear::new(784, 256);

pub fn with_seed( in_features: usize, out_features: usize, seed: Option<u64>, ) -> Self

Create a Linear layer with a specific random seed.

pub fn without_bias(in_features: usize, out_features: usize) -> Self

Create a Linear layer without bias.

Useful when followed by BatchNorm which has its own bias.

pub fn without_bias_with_seed( in_features: usize, out_features: usize, seed: Option<u64>, ) -> Self

Create a Linear layer without bias with a specific random seed.

pub fn in_features(&self) -> usize

Get the input feature dimension.

pub fn out_features(&self) -> usize

Get the output feature dimension.

pub fn has_bias(&self) -> bool

Check if this layer has a bias term.

pub fn set_weight(&mut self, weight: Tensor)

Set weight tensor from external data.

Used for loading pre-trained weights from SafeTensors or other formats. Automatically computes and caches the transposed weight for fast forward.

pub fn set_bias(&mut self, bias: Tensor)

Set bias tensor from external data.

Used for loading pre-trained weights.

pub fn placeholder(in_features: usize, out_features: usize) -> Self

Create a placeholder Linear layer with minimal memory allocation.

Used for lazy initialization when loading pre-trained weights. The placeholder uses 1-element tensors instead of full matrices, reducing memory from O(in*out) to O(1).

IMPORTANT: This layer will NOT work for inference until set_weight() is called with real weights.

pub fn weight(&self) -> &Tensor

Get reference to weight tensor.

pub fn bias(&self) -> Option<&Tensor>

Get reference to bias tensor if present.

pub fn is_ready(&self) -> bool

Check if this layer is ready for inference (weight_t is cached).

Returns false for placeholder layers that haven’t had set_weight() called. This is useful for verifying all layers are properly initialized before forward.

Trait Implementations

impl Debug for Linear

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

Formats the value using the given formatter.

impl Module for Linear

fn forward(&self, input: &Tensor) -> Tensor

Perform forward computation.

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

Get references to all learnable parameters.

fn parameters_mut(&mut self) -> Vec<&mut Tensor>

Get mutable references to all learnable parameters.

fn refresh_caches(&mut self)

Refresh any cached computations after parameters have been modified.

fn train(&mut self)

Set the module to training mode.

fn eval(&mut self)

Set the module to evaluation mode.

fn training(&self) -> bool

Check if the module is in training mode.

fn zero_grad(&mut self)

Zero out gradients for all parameters.

fn num_parameters(&self) -> usize

Get the number of learnable parameters.