blr-active: Active Learning Orchestration for Sensor Calibration

4-phase active learning workflow for noise-aware, precision-targeted sensor calibration.

Overview

blr-active implements a 4-phase active learning workflow for precision-targeted sensor calibration, built on top of the blr-core crate's Bayesian Linear Regression with Automatic Relevance Determination.

blr-core  (pure BLR+ARD math)
    └── blr-active  (this crate: active learning loop, precision tiers)

Features

✓ 4-Phase Calibration Workflow: Noise characterisation → exploration → active learning → completion
✓ Noise-Aware Precision Tiers: Map sensor noise floor to realistic precision goals (Low/Moderate/High/Max)
✓ 5 Active Learning Algorithms: Variance sampling, mutual information, goal-based, noise-floor detection
✓ Feasibility Assessment: Predicts whether a precision goal is achievable before collecting data
✓ Session History Export: JSON export of all measurements, precision records, and convergence data
✓ Zero panics on valid input: All errors returned as ALError

Installation

[dependencies]
blr-active = "0.1"
blr-core   = "0.1"  # required for BLR+ARD fitting

Quick Start: Noise Characterisation

use blr_active::{NoiseCalibrationSession, PrecisionLevel};
use blr_core::noise_estimation::estimate_noise_with_confidence;

// After an initial fit on 15 calibration points (beta from blr-core):
let noise_est = estimate_noise_with_confidence(25.0, 15, 6);
let mut noise_session = NoiseCalibrationSession::new(noise_est);

// Inspect precision tiers
for tier in noise_session.precision_tiers() {
    println!("{:?}: target std = {:.4}", tier.level, tier.target_std);
}

// Set precision goal
noise_session.set_goal(PrecisionLevel::Moderate).expect("feasible goal");

Quick Start: Active Learning Loop

use blr_active::{CalibrationSession, SessionConfig, IterationOutcome};

let config = SessionConfig::default();
let mut session = CalibrationSession::new(config);

// Seed with initial measurements
for (x, y) in initial_data {
    session.add_measurement(x, y);
}

loop {
    match session.next_iteration()? {
        IterationOutcome::RecommendNext { sample } => {
            let y = measure_sensor(sample.x);
            session.add_measurement(sample.x, y);
        }
        IterationOutcome::PrecisionMet { final_std } => {
            println!("Done! Precision std: {:.4}", final_std);
            break;
        }
        IterationOutcome::NoiseFloorHit => { println!("At noise floor."); break; }
        IterationOutcome::MaxIterationsReached => { break; }
    }
}

The 4-Phase Workflow

Phase	Component	Description
0	`NoiseCalibrationSession`	Noise characterisation → choose precision tier
1	`CalibrationSession`	Initial data collection across sensor range
2	`CalibrationSession::next_iteration()`	Active sample selection → converge
3	`export_history_json()`	Store calibration results and provenance

Precision Tiers

Tier	Factor × σ_noise	Est. samples	Use case
Low	5.0×	~10	Rapid commissioning
Moderate	3.0×	~25	Typical production calibration
High	1.5×	~60	Demanding applications
Max	1.0×	~200	Theoretical sensor limit

Algorithm Summary

Module	Algorithm	Description
`active_learning::variance`	Alg. 1	Posterior variance computation
`active_learning::acquisition`	Alg. 2	Variance-maximizing acquisition
`active_learning::precision`	Alg. 3	Precision assessment and goal checking
`active_learning::noise_floor`	Alg. 4	Noise floor (plateau) detection
`active_learning::orchestration`	Alg. 5	Synchronous calibration state machine

Examples

cargo run --example quick_start -p blr-active
cargo run --example multi_sensor_workflow -p blr-active

Compatibility

Rust: 1.70+
Platforms: Linux, macOS, Windows, WebAssembly (WASI Preview 2)
License: Apache 2.0

References

Settles, B. (2009). "Active Learning Literature Survey." UW-Madison TR 1648.
MacKay, D. J. C. (1992). "Information-Based Objective Functions for Active Data Selection." Neural Computation, 4(4), 590–604.
Tipping, M. E. (2001). "Sparse Bayesian Learning and the Relevance Vector Machine." JMLR, 1, 211–244.

License

Licensed under the Apache License, Version 2.0. See LICENSE for details.

Contributing

Contributions are welcome. Please ensure:

All tests pass: cargo test -p blr-active
Documentation is updated: cargo doc -p blr-active --no-deps --open
Code follows project conventions (see blr-core for style reference)

blr-active 0.1.0