Enum PruningSchedule 

pub enum PruningSchedule {
    OneShot {
        step: usize,
    },
    Gradual {
        start_step: usize,
        end_step: usize,
        initial_sparsity: f32,
        final_sparsity: f32,
        frequency: usize,
    },
    Cubic {
        start_step: usize,
        end_step: usize,
        final_sparsity: f32,
    },
}

Pruning schedule defining when sparsity increases during training.

Variants

• OneShot: All pruning happens at a single step
• Gradual: Linear interpolation between initial and final sparsity
• Cubic: Cubic polynomial schedule for smoother transitions

Example

use entrenar::prune::PruningSchedule;

// One-shot pruning at step 1000
let oneshot = PruningSchedule::OneShot { step: 1000 };
assert_eq!(oneshot.sparsity_at_step(500), 0.0);
assert_eq!(oneshot.sparsity_at_step(1000), 1.0);

// Gradual pruning from steps 100-1000
let gradual = PruningSchedule::Gradual {
    start_step: 100,
    end_step: 1000,
    initial_sparsity: 0.0,
    final_sparsity: 0.5,
    frequency: 10,
};
assert_eq!(gradual.sparsity_at_step(50), 0.0);
assert_eq!(gradual.sparsity_at_step(1000), 0.5);

Variants

OneShot

Prune once at specified step.

Fields

step: usize

Step at which to apply pruning.

Gradual

Gradually increase sparsity over steps with linear interpolation.

Fields

start_step: usize

Step to begin pruning.

end_step: usize

Step at which final sparsity is reached.

initial_sparsity: f32

Initial sparsity (typically 0.0).

final_sparsity: f32

Target final sparsity.

frequency: usize

Prune every N steps.

Cubic

Cubic sparsity schedule (Zhu & Gupta, 2017).

Formula: s_t = s_f * (1 - (1 - t/T)^3)

This provides faster initial pruning that slows as it approaches the target, giving the model more time to adapt.

Fields

start_step: usize

Step to begin pruning.

end_step: usize

Step at which final sparsity is reached.

final_sparsity: f32

Target final sparsity.

Implementations

impl PruningSchedule

pub fn sparsity_at_step(&self, step: usize) -> f32

Compute the target sparsity at a given training step.

Arguments

• step - Current training step

Returns

Target sparsity as a value between 0.0 and 1.0.

Panics

Does not panic. Returns bounded values for all inputs.

pub fn should_prune_at_step(&self, step: usize) -> bool

Check if pruning should be applied at this step based on frequency.

For OneShot, returns true only at the prune step. For Gradual, returns true at steps matching the frequency. For Cubic, returns true at every step in the pruning window.

Arguments

• step - Current training step

pub fn num_pruning_steps(&self) -> usize

Get the total number of pruning operations for this schedule.

Returns

Expected number of times pruning will be applied.

pub fn validate(&self) -> Result<(), String>

Check if the schedule is valid.

Errors

Returns an error message if the schedule is invalid.

pub fn is_complete(&self, step: usize) -> bool

Check if pruning has completed (current step is past the schedule).

Trait Implementations

impl Clone for PruningSchedule

fn clone(&self) -> PruningSchedule

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PruningSchedule

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

Formats the value using the given formatter.

impl Default for PruningSchedule

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for PruningSchedule

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for PruningSchedule

fn eq(&self, other: &PruningSchedule) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for PruningSchedule

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for PruningSchedule