Struct SpatialDropout 

pub struct SpatialDropout<A: Float> { /* private fields */ }

Spatial Dropout regularizer

Drops entire feature maps instead of individual units. This helps preserve spatial structure in convolutional neural networks.

Example

use scirs2_core::ndarray::Array4;
use optirs_core::regularizers::SpatialDropout;

let spatial_dropout = SpatialDropout::new(0.3).unwrap(); // 30% dropout rate

// 4D tensor (batch, channels, height, width)
let features = Array4::<f64>::ones((2, 3, 4, 4));

// During training - drops entire channels
let masked_features = spatial_dropout.apply(&features, true);

Implementations

impl<A: Float + Debug + ScalarOperand + Send + Sync> SpatialDropout<A>

pub fn new(dropprob: A) -> Result<Self>

Create a new SpatialDropout regularizer

Arguments

• dropprob - Probability of dropping each feature map (0.0 to 1.0)

pub fn with_feature_dim(self, dim: usize) -> Self

Set the dimension along which to drop features

pub fn apply<D>(&self, features: &Array<A, D>, training: bool) -> Array<A, D>

where D: Dimension + RemoveAxis,

Apply spatial dropout to a tensor

Trait Implementations

impl<A: Clone + Float> Clone for SpatialDropout<A>

fn clone(&self) -> SpatialDropout<A>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<A: Debug + Float> Debug for SpatialDropout<A>

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

Formats the value using the given formatter.

impl<A: Float + Debug + ScalarOperand + Send + Sync, D: Dimension + RemoveAxis + Send + Sync> Regularizer<A, D> for SpatialDropout<A>

fn apply(&self, params: &Array<A, D>, gradients: &mut Array<A, D>) -> Result<A>

Apply regularization to parameters and gradients

fn penalty(&self, params: &Array<A, D>) -> Result<A>

Compute the regularization penalty value