Struct IndexOptimizer 

pub struct IndexOptimizer { /* private fields */ }

Hyperparameter search engine for ANN index configurations.

Collects benchmark observations and provides:

• best-parameter selection for a given OptimizationTarget,
• Pareto-front extraction in the recall–QPS plane,
• a simple grid-expansion suggestion for the next candidate.

Implementations

impl IndexOptimizer

pub fn new(index_type: IndexType, target: OptimizationTarget) -> Self

Create a new optimiser for index_type that pursues target.

pub fn add_benchmark(&mut self, point: BenchmarkPoint)

Record a new benchmark observation.

pub fn benchmark_count(&self) -> usize

Number of recorded benchmark observations.

pub fn best_params(&self) -> Option<&BenchmarkPoint>

Return the benchmark point with the highest score under the current target, or None if no benchmarks have been recorded.

pub fn pareto_front(&self) -> Vec<&BenchmarkPoint>

Extract the Pareto-optimal front in the (recall_at_10, qps) plane.

A point a dominates b when a.recall_at_10 >= b.recall_at_10 and a.qps >= b.qps with at least one strict inequality. The returned vector contains only non-dominated points, sorted by descending recall.

pub fn suggest_next_params(&self) -> Option<IndexParams>

Suggest the next parameter configuration to benchmark.

The strategy is a simple grid expansion: take the best observed configuration and propose a neighbouring point by incrementing the most impactful parameter by one step. Returns None when no benchmarks exist or the index type is Flat (no free parameters).

pub fn index_type(&self) -> &IndexType

Reference to the index type being optimised.

pub fn benchmarks(&self) -> &[BenchmarkPoint]

All recorded benchmark points.

pub fn clear(&mut self)

Clear all recorded benchmarks (useful for re-runs).

pub fn ranked_benchmarks(&self) -> Vec<&BenchmarkPoint>

Return all benchmarks sorted by score under the current target (descending).

pub fn score_of(&self, point: &BenchmarkPoint) -> f64

Return the score of the given benchmark point under the current target.

pub fn filter_by_recall(&self, min_recall: f64) -> Vec<&BenchmarkPoint>

Return benchmarks that achieve at least min_recall recall@10.

pub fn filter_by_qps(&self, min_qps: f64) -> Vec<&BenchmarkPoint>

Return benchmarks that achieve at least min_qps queries per second.

pub fn recall_stats(&self) -> Option<RecallStats>

Compute summary statistics (min/max/mean) over recall@10 values.

pub fn group_by_variant(&self) -> HashMap<&'static str, Vec<&BenchmarkPoint>>

Group benchmarks by their parameter variant (Hnsw vs Ivf).