Struct SpatialLengthScaleOptimizationTiming 

pub struct SpatialLengthScaleOptimizationTiming {

    pub log_kappa_dim: usize,
    pub cost_calls: usize,
    pub cost_total_s: f64,
    pub eval_calls: usize,
    pub eval_total_s: f64,
    pub efs_calls: usize,
    pub efs_total_s: f64,
    pub slow_path_resets: u64,
    pub design_revision_delta: u64,
    pub nfree_skip_row_touches: u64,
    pub nfree_miss_shape: u64,
    pub nfree_miss_value: u64,
    pub nfree_miss_gradient: u64,
    pub nfree_miss_penalty: u64,
    pub nfree_miss_revision: u64,
    pub nfree_miss_second_order: u64,
    pub nfree_miss_other: u64,
    pub exact_polish_ran: bool,
    pub polish_slow_path_resets: u64,
    pub polish_nfree_skip_row_touches: u64,
    pub optim_total_s: f64,
}

Fields

log_kappa_dim: usize

cost_calls: usize

cost_total_s: f64

eval_calls: usize

eval_total_s: f64

efs_calls: usize

efs_total_s: f64

slow_path_resets: u64

design_revision_delta: u64

nfree_skip_row_touches: u64

#1868 deterministic n-independence instrument: the number of length-n row-element touches the Gaussian zero-iteration inner synthesis performed on the #1033 n-free κ-trial skip path during this κ-optimisation phase (excludes the one-time priming eval). The #1033 architectural invariant requires each in-window trial to touch only k×k objects, so this MUST NOT scale with n. A value that grows with n is exactly the #1868 O(n)-per-callback regression. This replaces the old noisy wall-clock per-callback ratio with an exact, millisecond-fast integer gate.

nfree_miss_shape: u64

nfree_miss_value: u64

nfree_miss_gradient: u64

nfree_miss_penalty: u64

nfree_miss_revision: u64

nfree_miss_second_order: u64

nfree_miss_other: u64

exact_polish_ran: bool

Whether begin_exact_polish retired the #1033b n-free ψ-Gram surrogate and the optimizer continued on the exact streamed criterion (gam#2760).

Every counter above is a statement about the SEARCH and stops at this boundary — “an in-window hyperparameter TRIAL touches only k×k objects” is a claim about trials, and the polish is not a trial phase. The two fields below carry the polish’s own O(n) cost so it is published rather than either hidden or charged to the search.

polish_slow_path_resets: u64

slow_path_resets accrued AFTER the exact-polish boundary. Bounded by the polish’s own iteration budget, so n-independent; every one of them is intended, because the polish exists precisely to leave the n-free lane.

polish_nfree_skip_row_touches: u64

nfree_skip_row_touches accrued after the exact-polish boundary. The surrogate is gone by then, so no evaluation can take the skip path and this must stay 0 — a nonzero value would mean the retirement did not take.

optim_total_s: f64

Implementations

impl SpatialLengthScaleOptimizationTiming

pub fn trial_total_s(self) -> f64

Trait Implementations

impl Clone for SpatialLengthScaleOptimizationTiming

fn clone(&self) -> SpatialLengthScaleOptimizationTiming

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for SpatialLengthScaleOptimizationTiming

impl Debug for SpatialLengthScaleOptimizationTiming

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

Formats the value using the given formatter.

impl Default for SpatialLengthScaleOptimizationTiming

fn default() -> SpatialLengthScaleOptimizationTiming

Returns the “default value” for a type.