Module evaluation 

Evaluation metrics and infrastructure for ANN benchmarks.

Standard metrics used in ANN literature:

Metric	Formula	Interpretation
Recall@K	|approx ∩ true| / K	Fraction of true neighbors found
Precision@K	|approx ∩ true| / |approx|	Same as recall when |approx| = K
MRR	1 / rank_of_first_true	Reciprocal of first relevant result
QPS	queries / seconds	Throughput

Standard Benchmark Datasets

Dataset	Size	Dim	Distance	Source
SIFT-1M	1M	128	L2	INRIA Texmex
GIST-1M	1M	960	L2	INRIA Texmex
GloVe-100	1.2M	100	Angular	Stanford NLP
Fashion-MNIST	60K	784	L2	Zalando
Deep1B	1B	96	Angular	Yandex

Reference: https://ann-benchmarks.com/

Re-exports

pub use crate::distance::angular_distance;

pub use crate::distance::cosine_distance;

pub use crate::distance::inner_product_distance;

pub use crate::distance::l2_distance;

pub use crate::distance::normalize;

pub use crate::distance::DistanceMetric;

Structs

EvalDataset

A dataset with ground truth for evaluation.

EvalResults

Evaluation results for a single run.

Functions

compute_ground_truth

Compute ground truth for a dataset.

evaluate

Evaluate an algorithm on a dataset.

generate_clustered_dataset

Generate a clustered dataset (more realistic).

generate_normalized_clustered_dataset

Generate a normalized dataset (for cosine/angular metrics).

generate_uniform_dataset

Generate a synthetic dataset with uniform random vectors.

mrr

Compute mean reciprocal rank for a single query.

recall_at_k

Compute recall@k for a single query.