Struct SparseDictConfig 

pub struct SparseDictConfig {
    pub n_atoms: usize,
    pub active: usize,
    pub minibatch: usize,
    pub max_epochs: usize,
    pub score_tile: usize,
    pub code_ridge: f32,
    pub decoder_ridge: f32,
    pub tolerance: f64,
}

Shared (NOT per-atom) hyper-parameters for the collapsed linear lane.

The whole point of the sparse trainer is that K is too large to carry a per-atom smoothing parameter / Newton state; every knob here is a single scalar shared across the entire dictionary.

Fields

n_atoms: usize

Dictionary width K (number of atoms).

active: usize

Active budget s: how many atoms may fire per row (top_s). This is the shared routing-sparsity hyper-parameter.

minibatch: usize

Minibatch size (rows per route→code→accumulate step). The decoder is refreshed once per full epoch from the accumulated sparse normal equations, so this only bounds peak working set, not the solution.

max_epochs: usize

Number of full passes over the data.

score_tile: usize

Column tile width used when scoring rows against the dictionary. Score tiles of shape minibatch × tile are formed and discarded; the N×K score matrix is never materialised.

code_ridge: f32

Shared ridge on the per-row active-set code solve (Tikhonov on the s×s Gram). Identifies the codes when active atoms are collinear.

decoder_ridge: f32

Shared ridge on the per-atom decoder refresh (method-of-optimal -directions normal equations). Keeps a thinly-used atom well posed.

tolerance: f64

Relative explained-variance improvement below which training stops.

Implementations

impl SparseDictConfig

pub fn new(n_atoms: usize) -> Self

Construct a config for a K-atom dictionary, leaving every other knob at its shared default.

Trait Implementations

impl Clone for SparseDictConfig

fn clone(&self) -> SparseDictConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for SparseDictConfig

impl Debug for SparseDictConfig

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

Formats the value using the given formatter.

impl Default for SparseDictConfig

fn default() -> Self

Returns the “default value” for a type.