Struct LotteryTicketState 

pub struct LotteryTicketState {
    pub initial_weights: Vec<WeightTensor>,
    pub masks: Vec<PruningMask>,
    pub iteration: usize,
    pub sparsity_history: Vec<f32>,
    pub enabled: bool,
}

State for Lottery Ticket Hypothesis rewinding

The Lottery Ticket Hypothesis (Frankle & Carlin, 2019) suggests that randomly initialized networks contain sparse subnetworks (“winning tickets”) that can achieve comparable accuracy when trained in isolation.

Fields

initial_weights: Vec<WeightTensor>

Initial weights before any training (for rewinding)

masks: Vec<PruningMask>

Current pruning masks learned through training

iteration: usize

Current pruning iteration

sparsity_history: Vec<f32>

Sparsity at each iteration

enabled: bool

Whether rewinding is enabled

Implementations

impl LotteryTicketState

pub fn new(initial_weights: Vec<WeightTensor>) -> Self

Creates a new lottery ticket state

pub fn rewind(&self) -> Vec<WeightTensor>

Rewinds weights to initial values while applying current masks

Returns the rewound weights with masks applied

pub fn update_masks(&mut self, new_masks: Vec<PruningMask>, sparsity: f32)

Updates masks after a pruning iteration

pub fn current_sparsity(&self) -> f32

Returns the current overall sparsity

Trait Implementations

impl Clone for LotteryTicketState

fn clone(&self) -> LotteryTicketState

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LotteryTicketState

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

Formats the value using the given formatter.