Quality Metrics

Overview

Evaluate model and search quality using various metrics.

Recall@K

Fraction of relevant items in top K results:

-- Calculate Recall@K
SELECT recall_at_k(
    ARRAY[1, 2, 3],      -- retrieved items
    ARRAY[1, 2, 5, 6],   -- relevant items
    5                    -- K
);

Precision@K

Fraction of retrieved items that are relevant:

-- Calculate Precision@K
SELECT precision_at_k(
    ARRAY[1, 2, 3],
    ARRAY[1, 2, 5],
    5
);

F1@K

Harmonic mean of Precision@K and Recall@K:

-- Calculate F1@K
SELECT f1_at_k(
    ARRAY[1, 2, 3],
    ARRAY[1, 2, 5],
    5
);

MRR (Mean Reciprocal Rank)

Average reciprocal rank of first relevant result:

-- Calculate MRR
SELECT mean_reciprocal_rank(
    ARRAY[
        ARRAY[1, 2, 3],
        ARRAY[5, 1, 2]
    ],
    ARRAY[1, 1]  -- relevant items per query
);

Davies-Bouldin Index

Clustering quality metric (lower is better):

-- Calculate Davies-Bouldin Index
SELECT davies_bouldin_index(
    'data_table',
    'features',
    'cluster_label'
);

Learn More

For detailed documentation on all quality metrics, choosing appropriate metrics, benchmarking, and interpretation, visit: Quality Metrics Documentation

Related Topics

• Clustering - Evaluate clustering quality
• Vector Search - Evaluate search quality