Crate epsilon_engine

Expand description

Pillar: III. PACR field: Γ (Cognitive Complexity).

Full CSSR ε-machine inference producing CognitiveSplit (S_T, H_T) with bootstrap confidence intervals.

§Quick start

use epsilon_engine::{Config, infer};

let symbols: Vec<u8> = vec![0, 1, 0, 1, 0, 1]; // tiny example
let cfg = Config::default();
let result = infer(&symbols, cfg);
assert!(result.cognitive_split.statistical_complexity.point >= 0.0);

§CSSR algorithm summary

See cssr.rs for the full algorithm description. Briefly: suffix-tree statistics → KS homogeneity tests → state splitting → merge pass → compact → stationary π → (C_μ, h_μ) with B=200 bootstrap CIs.

§Memory budget

Suffix table entries: at most min(N, |A|^L) per depth level. For N=100 000, L=12, |A|=8: unique depth-L histories ≤ 100 000. Total entries across all depths ≤ N × L = 1 200 000. Each entry: Vec of |A|=8 elements ≈ 32 B → ~38 MiB worst-case, well under the 200 MiB budget.

Re-exports§

pub use symbolize::WordSymbolizer;

Modules§

bootstrap_backend: Pillar: III. PACR field: Γ.
complexity: Pillar: III. PACR field: Γ.
cssr: Pillar: III. PACR field: Γ.
quick_screen: Pillar: III. PACR field: Γ.
symbolize: Pillar: III. PACR field: Γ (symbol alphabet for ε-machine construction).

Structs§

Config: Configuration for the ε-machine inference.
InferResult: Full result of ε-machine inference.

Functions§

empirical_entropy_rate: Compute empirical entropy rate of a symbol sequence.
infer: Infer the ε-machine from a discrete symbol sequence.
infer_fast: Infer without bootstrap CI (faster; Estimate::exact point estimates only).