Struct AdaptiveState 

pub struct AdaptiveState {
    pub baseline_samples: Vec<u64>,
    pub sample_samples: Vec<u64>,
    pub previous_posterior: Option<Posterior>,
    pub recent_kl_divergences: VecDeque<f64>,
    pub batch_count: usize,
    /* private fields */
}

State maintained during adaptive sampling loop.

This struct accumulates timing samples and tracks the evolution of the posterior distribution across batches, enabling quality gate checks like KL divergence monitoring.

Also maintains online statistics (mean, variance, lag-1 autocorrelation) for condition drift detection between calibration and post-test phases.

Time Tracking

This struct does NOT track elapsed time internally. The caller must provide elapsed_secs to functions that need it (e.g., quality gate checks). This allows use in no_std environments where std::time::Instant is unavailable.

Fields

baseline_samples: Vec<u64>

Baseline class timing samples (in cycles/ticks/native units).

sample_samples: Vec<u64>

Sample class timing samples (in cycles/ticks/native units).

previous_posterior: Option<Posterior>

Previous posterior for KL divergence tracking. None until we have at least one posterior computed.

recent_kl_divergences: VecDeque<f64>

Recent KL divergences (last 5 batches) for learning rate monitoring. If sum of recent KL < 0.001, learning has stalled.

batch_count: usize

Number of batches collected so far.

Implementations

impl AdaptiveState

pub fn new() -> Self

Create a new empty adaptive state.

pub fn with_capacity(expected_samples: usize) -> Self

Create a new adaptive state with pre-allocated capacity.

pub fn n_total(&self) -> usize

Get the total number of samples per class.

pub fn add_batch(&mut self, baseline: Vec<u64>, sample: Vec<u64>)

Add a batch of samples to the state.

Both baseline and sample vectors should have the same length. Note: This method does not track online statistics since ns_per_tick is not known. Use add_batch_with_conversion if you need drift detection.

pub fn add_batch_with_conversion( &mut self, baseline: Vec<u64>, sample: Vec<u64>, ns_per_tick: f64, )

Add a batch of samples and track online statistics for drift detection.

Both baseline and sample vectors should have the same length.

Arguments

• baseline - Baseline class timing samples (in native units)
• sample - Sample class timing samples (in native units)
• ns_per_tick - Conversion factor from native units to nanoseconds

pub fn update_kl(&mut self, kl: f64)

Update KL divergence history with a new value.

Maintains a sliding window of the last 5 KL divergences for learning rate monitoring.

pub fn recent_kl_sum(&self) -> f64

Get the sum of recent KL divergences.

Used to detect learning stall (sum < 0.001 indicates posterior has stopped updating despite new data).

pub fn has_kl_history(&self) -> bool

Check if we have enough KL history for learning rate assessment.

pub fn update_posterior(&mut self, new_posterior: Posterior) -> f64

Update the posterior and track KL divergence.

Returns the KL divergence from the previous posterior, or 0.0 if this is the first posterior.

pub fn current_posterior(&self) -> Option<&Posterior>

Get the current posterior, if computed.

pub fn baseline_ns(&self, ns_per_tick: f64) -> Vec<f64>

Convert baseline samples to f64 nanoseconds.

pub fn sample_ns(&self, ns_per_tick: f64) -> Vec<f64>

Convert sample samples to f64 nanoseconds.

pub fn baseline_stats(&self) -> Option<StatsSnapshot>

Get the current online statistics for the baseline class.

Returns None if no samples have been added with conversion tracking.

pub fn sample_stats(&self) -> Option<StatsSnapshot>

Get the current online statistics for the sample class.

Returns None if no samples have been added with conversion tracking.

pub fn get_stats_snapshot(&self) -> Option<CalibrationSnapshot>

Get a CalibrationSnapshot from the current online statistics.

Returns None if insufficient samples have been tracked.

pub fn has_stats_tracking(&self) -> bool

Check if online statistics are being tracked.

Returns true if add_batch_with_conversion has been used.

pub fn reset(&mut self)

Reset the state for a new test run.

Clears all samples, posteriors, and statistics while preserving capacity.

Trait Implementations

impl Default for AdaptiveState

fn default() -> Self

Returns the “default value” for a type.