Struct ProjectionConfig 

pub struct ProjectionConfig {
    pub rank: usize,
    pub lambda: f64,
    pub delta: f64,
    pub warmup: usize,
    pub seed: u64,
    pub supervised_lr: f64,
}

Configuration for online projection learning.

Controls the PAST subspace tracker and the warmup period for the normalizer. Use ProjectionConfig::builder() for ergonomic construction with validation.

Defaults

Parameter	Default	Description
rank	8	Projection rank (output dimension)
lambda	0.9999	PAST forgetting factor (half-life ~6931 samples)
delta	100.0	Initial P diagonal scaling
warmup	200	Warmup samples (normalizer only, no PAST updates)
seed	42	RNG seed for Xavier initialization
supervised_lr	0.001	Learning rate for supervised projection gradient

Fields

rank: usize

Projection rank (output dimension). Default: 8.

lambda: f64

PAST forgetting factor (half-life ~6931 samples). Default: 0.9999.

delta: f64

Initial P diagonal scaling. Default: 100.0.

warmup: usize

Warmup samples (pass raw features before updating PAST). Default: 200.

seed: u64

RNG seed. Default: 42.

supervised_lr: f64

Learning rate for supervised projection gradient. Default: 0.01.

When the inner model exposes readout weights, this learning rate controls how fast W adapts to prediction-relevant directions. Decoupled from lambda because the projection learning timescale and forgetting timescale serve different purposes.

Implementations

impl ProjectionConfig

pub fn builder() -> ProjectionConfigBuilder

Start building a ProjectionConfig with default values.

Trait Implementations

impl Clone for ProjectionConfig

fn clone(&self) -> ProjectionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ProjectionConfig

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

Formats the value using the given formatter.

impl Default for ProjectionConfig

fn default() -> Self

Returns the “default value” for a type.