Enum RateBackend 

#[non_exhaustive]
pub enum RateBackend {
    RosaPlus {
        max_order: i64,
    },
    Match {
        hash_bits: usize,
        min_len: usize,
        max_len: usize,
        base_mix: f64,
        confidence_scale: f64,
    },
    SparseMatch {
        hash_bits: usize,
        min_len: usize,
        max_len: usize,
        gap_min: usize,
        gap_max: usize,
        base_mix: f64,
        confidence_scale: f64,
    },
    Ppmd {
        order: usize,
        memory_mb: usize,
    },
    Sequitur {
        context_bytes: usize,
    },
    MambaMethod {
        method: MethodSpec,
    },
    Rwkv7Method {
        method: MethodSpec,
    },
    Zpaq {
        method: ZpaqMethodSpec,
    },
    Mixture {
        spec: Arc<MixtureSpec>,
    },
    Particle {
        spec: Arc<ParticleSpec>,
    },
    Calibrated {
        spec: Arc<CalibratedSpec>,
    },
    Ctw {
        depth: usize,
    },
    FacCtw {
        base_depth: usize,
        num_percept_bits: usize,
        encoding_bits: usize,
        msb_first: Option<bool>,
    },
}

Core predictive model class used by the library.

RateBackend is the shared model class behind entropy-rate estimation, rate-coded compression, generation, and the world-model interface used by MC-AIXI/AIQI planners.

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

RosaPlus

ROSA+ suffix-automaton estimator.

max_order < 0 enables ROSA’s adaptive order selection over the full suffix automaton; max_order >= 0 caps the predictive context length.

Fields

max_order: i64

Maximum context length used by the ROSA predictor.

< 0 means “adaptive / full SAM”; >= 0 is a fixed cap.

Match

Local contiguous match predictor.

Fields

hash_bits: usize

Number of retained hash bits for suffix lookup.

min_len: usize

Minimum repeat length required before predicting.

max_len: usize

Maximum repeat length used for confidence scaling.

base_mix: f64

Residual probability mass left for non-match symbols.

confidence_scale: f64

Confidence multiplier applied to short-match tapering.

SparseMatch

Sparse/gapped local match predictor.

Fields

hash_bits: usize

Number of retained hash bits for spaced-suffix lookup.

min_len: usize

Minimum spaced repeat length required before predicting.

max_len: usize

Maximum spaced repeat length used for confidence scaling.

gap_min: usize

Minimum gap between matched bytes.

gap_max: usize

Maximum gap between matched bytes.

base_mix: f64

Residual probability mass left for non-match symbols.

confidence_scale: f64

Confidence multiplier applied to short-match tapering.

Ppmd

Pure-Rust bounded-memory PPMD-style model.

Fields

order: usize

Maximum context order.

memory_mb: usize

Approximate memory budget in MiB.

Sequitur

Exact online Sequitur grammar backend with byte-level predictive readout.

Fields

context_bytes: usize

Maximum number of terminal bytes retained per grammar-derived context.

MambaMethod

Typed Mamba method specification.

Fields

method: MethodSpec

Mamba method specification.

Rwkv7Method

Typed RWKV7 method specification.

Fields

method: MethodSpec

RWKV7 method specification.

Zpaq

ZPAQ compression-based rate model (streamable methods only).

Fields

method: ZpaqMethodSpec

Typed ZPAQ method specification.

Mixture

Online mixture over RateBackend experts.

Bayes, Switching, and Convex follow “On Ensemble Techniques for AIXI Approximation”; FadingBayes, Mdl, and Neural are repository extensions.

Fields

spec: Arc<MixtureSpec>

Mixture expert/runtime specification.

Particle

Particle-latent filter ensemble.

Fields

spec: Arc<ParticleSpec>

Particle filter specification.

Calibrated

Calibrated wrapper over another bytewise backend.

Fields

spec: Arc<CalibratedSpec>

Calibration specification.

Ctw

Action-Conditional CTW (single context tree).

Fields

depth: usize

Context tree depth.

FacCtw

Factorized Action-Conditional CTW (k trees for k-bit percepts).

Fields

base_depth: usize

Base context depth.

num_percept_bits: usize

Planner/percept bit cardinality for AIXI-style consumers.

Rate-backend byte execution uses encoding_bits as the symbol decomposition width. num_percept_bits is retained for controller specs whose percept cardinality can differ from byte encoding width.

encoding_bits: usize

Encoding width in bits for rate execution.

Valid compiled FAC-CTW specs require 1..=8.

msb_first: Option<bool>

Optional bit order for decomposing symbols.

None keeps the compatibility default: 8-bit byte symbols use MSB-first ordering, while non-byte widths retain the legacy LSB-first FAC-CTW convention. Some(true) selects MSB-first explicitly; Some(false) selects legacy LSB-first explicitly.

Implementations

impl RateBackend

pub fn try_default() -> InfotheoryResult<Self>

Returns the current build’s implicit default rate backend.

pub fn validate_in( &self, env: &SpecEnvironment, ) -> SpecResult<ValidatedRateBackend>

Validate and canonicalize this backend in the provided compilation environment.

pub fn validate(&self) -> SpecResult<ValidatedRateBackend>

Validate and canonicalize this backend using the default environment.

pub fn compile_in( &self, env: &SpecEnvironment, ) -> SpecResult<CompiledRateBackend>

Validate and compile this backend in the provided compilation environment.

pub fn compile(&self) -> SpecResult<CompiledRateBackend>

Validate and compile this backend using the default environment.

Trait Implementations

impl CanonicalJson for RateBackend

fn to_canonical_json_value(&self) -> SpecResult<Value>

Serialize this value into canonical JSON value form.

fn to_canonical_json(&self) -> SpecResult<String>

Serialize this value into deterministic canonical JSON text.

impl Clone for RateBackend

fn clone(&self) -> RateBackend

Returns a duplicate of the value.

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

Performs copy-assignment from source.