Struct RealtimeAdaptationConfig 

pub struct RealtimeAdaptationConfig<T>
where
    T: Float + Debug + Send + Sync + 'static,
{

    pub loss_detector: DriftDetectorConfig<T>,
    pub grad_detector: DriftDetectorConfig<T>,
    pub initial_lr: T,
    pub lr_min: T,
    pub lr_max: T,
    pub lr_decay: T,
    pub lr_growth: T,
    pub initial_momentum: T,
    pub mom_min: T,
    pub mom_max: T,
    pub mom_damp: T,
    pub plateau_patience: usize,
    pub plateau_tol: T,
    pub cooldown_steps: usize,
    pub grad_norm_ceiling_factor: T,
}

Configuration for the RealtimeAdaptationController.

Fields

loss_detector: DriftDetectorConfig<T>

Detector configuration for the loss stream.

grad_detector: DriftDetectorConfig<T>

Detector configuration for the gradient-norm stream.

initial_lr: T

Initial learning rate.

lr_min: T

Lower clamp on the learning rate.

lr_max: T

Upper clamp on the learning rate.

lr_decay: T

Multiplicative factor < 1 applied to decrease the learning rate.

lr_growth: T

Multiplicative factor > 1 applied to increase the learning rate.

initial_momentum: T

Initial momentum.

mom_min: T

Lower clamp on momentum.

mom_max: T

Upper clamp on momentum.

mom_damp: T

Multiplicative factor <= 1 used to damp momentum on divergence.

plateau_patience: usize

Number of consecutive low-improvement loss steps that constitute a plateau.

plateau_tol: T

Relative-improvement tolerance below which a loss step counts as “no progress” for plateau detection.

cooldown_steps: usize

Number of steps after any adjustment during which no new adjustment is made.

grad_norm_ceiling_factor: T

Multiplier on the gradient-norm reference mean above which the gradient is considered “unbounded” (suppressing learning-rate increases).

Implementations

impl<T> RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static,

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

Validate the controller configuration and both detector configurations.

Errors

Returns OptimError::InvalidConfig describing the first malformed field encountered.

Trait Implementations

impl<T> Clone for RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static + Clone,

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T> Copy for RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static + Copy,

impl<T> Debug for RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static + Debug,

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

Formats the value using the given formatter.

impl<T> Default for RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static,

fn default() -> Self

Returns the “default value” for a type.

impl<T> PartialEq for RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static + PartialEq,

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

Equality operator ==.

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

Inequality operator !=.

impl<T> StructuralPartialEq for RealtimeAdaptationConfig<T>

where T: Float + Debug + Send + Sync + 'static + PartialEq,