Struct AiqiConfig 

pub struct AiqiConfig {

    pub algorithm: String,
    pub ct_depth: usize,
    pub observation_bits: usize,
    pub observation_stream_len: usize,
    pub reward_bits: usize,
    pub agent_actions: usize,
    pub min_reward: Reward,
    pub max_reward: Reward,
    pub reward_offset: Reward,
    pub discount_gamma: f64,
    pub return_horizon: usize,
    pub return_bins: usize,
    pub augmentation_period: usize,
    pub history_prune_keep_steps: Option<usize>,
    pub baseline_exploration: f64,
    pub random_seed: Option<u64>,
    pub rate_backend: Option<RateBackend>,
    pub rate_backend_max_order: i64,
    pub rwkv_model_path: Option<String>,
    pub rosa_max_order: Option<i64>,
    pub zpaq_method: Option<String>,
}

Configuration parameters for an AIQI agent.

Fields

algorithm: String

Predictive backend.

• ac-ctw / ctw / ctw-context-tree: paper AIQI-CTW path.
• fac-ctw: factorized CTW extension.
• rosa / rwkv: pluggable predictor extensions.
• zpaq: intentionally unsupported for AIQI strict conditioning.

ct_depth: usize

Context depth for CTW/FAC-CTW backends.

observation_bits: usize

Number of bits used to encode observations.

observation_stream_len: usize

Number of observation symbols per environment step.

reward_bits: usize

Number of bits used to encode rewards.

agent_actions: usize

Number of valid actions.

min_reward: Reward

Minimum possible environment reward.

max_reward: Reward

Maximum possible environment reward.

reward_offset: Reward

Offset applied before encoding reward bits.

discount_gamma: f64

Discount factor used when constructing H-step returns.

return_horizon: usize

Return horizon H.

return_bins: usize

Number of discretization bins M for returns.

augmentation_period: usize

Augmentation period N (must satisfy N >= H).

history_prune_keep_steps: Option<usize>

Optional history retention knob for bounded memory growth.

• None: keep full history (default behavior, no pruning).
• Some(k): keep at least the most recent k steps, while also preserving all steps still required for exact return construction and deferred phase-model advancement.

baseline_exploration: f64

Baseline epsilon-greedy exploration probability tau.

random_seed: Option<u64>

Optional deterministic RNG seed for action selection/exploration.

When None, a fresh runtime-derived seed is used.

rate_backend: Option<RateBackend>

Optional generic rate backend.

When set, this takes precedence over algorithm and routes AIQI prediction through the shared RateBackend abstraction.

rate_backend_max_order: i64

Max-order hint for rate_backend constructors that use it (for example ROSA).

rwkv_model_path: Option<String>

Optional RWKV model path.

Required only when selecting algorithm="rwkv" and no rate_backend override is configured.

rosa_max_order: Option<i64>

Optional ROSA max order.

zpaq_method: Option<String>

Optional ZPAQ method string.

Implementations

impl AiqiConfig

pub fn validate(&self) -> Result<(), String>

Validate configuration constraints.

Trait Implementations

impl Clone for AiqiConfig

fn clone(&self) -> AiqiConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.