Struct LinearStochasticDepth 

pub struct LinearStochasticDepth {
    pub num_layers: usize,
    pub drop_prob_min: f64,
    pub drop_prob_max: f64,
}

Linear stochastic depth scheduler.

Linearly increases drop probability from min to max across network depth. This is the standard approach used in most deep networks:

• Shallow layers: low drop probability (more stable)
• Deep layers: high drop probability (more regularization)

Example

use tensorlogic_train::LinearStochasticDepth;

// 10 layers, drop_prob from 0.0 to 0.3
let scheduler = LinearStochasticDepth::new(10, 0.0, 0.3).unwrap();

// Get drop probability for layer 5
let drop_prob_5 = scheduler.get_drop_prob(5);
assert!((drop_prob_5 - 0.15).abs() < 1e-6); // Halfway point

Fields

num_layers: usize

Total number of layers/blocks.

drop_prob_min: f64

Minimum drop probability (first layer).

drop_prob_max: f64

Maximum drop probability (last layer).

Implementations

impl LinearStochasticDepth

pub fn new( num_layers: usize, drop_prob_min: f64, drop_prob_max: f64, ) -> TrainResult<Self>

Create a new linear stochastic depth scheduler.

Arguments

• num_layers - Total number of layers in the network
• drop_prob_min - Drop probability for first layer (usually 0.0)
• drop_prob_max - Drop probability for last layer (e.g., 0.1-0.5)

Returns

A new scheduler or error if parameters are invalid.

pub fn get_drop_prob(&self, layer_idx: usize) -> f64

Get drop probability for a specific layer.

Uses linear interpolation:

drop_prob(i) = drop_prob_min + (drop_prob_max - drop_prob_min) * i / (L-1)

Arguments

• layer_idx - Layer index (0 to num_layers-1)

Returns

Drop probability for this layer.

pub fn create_drop_paths(&self) -> TrainResult<Vec<DropPath>>

Create DropPath instances for all layers.

Returns

Vector of DropPath instances with linearly increasing drop probabilities.

Trait Implementations

impl Clone for LinearStochasticDepth

fn clone(&self) -> LinearStochasticDepth

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LinearStochasticDepth

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

Formats the value using the given formatter.