Struct DropPath 

pub struct DropPath {
    pub drop_prob: f64,
    /* private fields */
}

DropPath (Stochastic Depth) regularization.

Randomly drops entire paths in residual networks during training. At test time, paths are always kept but scaled by (1 - drop_prob).

Example

use tensorlogic_train::DropPath;
use scirs2_core::ndarray::Array2;
use scirs2_core::random::{StdRng, SeedableRng};

let drop_path = DropPath::new(0.1).unwrap(); // 10% drop probability
let mut rng = StdRng::seed_from_u64(42);

let residual = Array2::ones((4, 8));

// Training mode: randomly drop
let output = drop_path.apply(&residual.view(), true, &mut rng).unwrap();

// Inference mode: scale by keep probability
let output_test = drop_path.apply(&residual.view(), false, &mut rng).unwrap();

Fields

drop_prob: f64

Probability of dropping the path (0.0 to 1.0).

Implementations

impl DropPath

pub fn new(drop_prob: f64) -> TrainResult<Self>

Create a new DropPath regularizer.

Arguments

• drop_prob - Probability of dropping the path (0.0 to 1.0)

Returns

A new DropPath instance or error if drop_prob is invalid.

pub fn apply( &self, path: &ArrayView2<'_, f64>, training: bool, rng: &mut StdRng, ) -> TrainResult<Array2<f64>>

Apply DropPath to a residual path.

Arguments

• path - The residual path to potentially drop
• training - Whether in training mode (drop) or inference mode (scale)
• rng - Random number generator

Returns

Transformed path (either dropped or scaled)

pub fn apply_batch( &self, paths: &ArrayView2<'_, f64>, training: bool, rng: &mut StdRng, ) -> TrainResult<Array2<f64>>

Apply DropPath to a batch of paths.

Each sample in the batch is independently dropped with probability drop_prob.

Arguments

• paths - Batch of residual paths (batch_size × features)
• training - Whether in training mode
• rng - Random number generator

Returns

Batch with randomly dropped paths

pub fn keep_probability(&self) -> f64

Get the keep probability.

pub fn set_drop_prob(&mut self, drop_prob: f64) -> TrainResult<()>

Set new drop probability.

Trait Implementations

impl Clone for DropPath

fn clone(&self) -> DropPath

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DropPath

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

Formats the value using the given formatter.