📏 Measuring Retrieval Performance

Evaluate retrieval quality with standard metrics.

Metrics Overview

Metric	Description
Recall@k	Fraction of relevant docs in top-k
P@k	Precision at k
MRR	Mean Reciprocal Rank
Top-k Accuracy	Whether answer exists in top-k

Usage

from rankify.metrics.metrics import Metrics

metrics = Metrics(documents)
results = metrics.calculate_retrieval_metrics(
    ks=[1, 5, 10, 20, 50, 100],
    use_reordered=False  # Original retrieval order
)

print(f"Recall@5: {results['Recall@5']:.4f}")
print(f"Recall@10: {results['Recall@10']:.4f}")
print(f"MRR: {results['MRR']:.4f}")

Interpreting Results

• Recall@k: Higher is better, measures coverage
• MRR: Measures where first relevant result appears
• P@k: Precision among top-k results