Retrieval Metrics

Metrics for evaluating document retrieval quality.

Available Metrics

Metric	Measures	When to Use
Recall@k	Coverage	Ensure all relevant docs found
Full Recall@k	Complete coverage	All evidence groups must be retrieved
Precision@k	Relevance	Minimize irrelevant results
F1@k	Balance	Trade-off recall and precision
NDCG@k	Ranking	Order matters
MRR	First hit	Single answer tasks
MAP	Overall quality	Comprehensive evaluation

Common Parameters

All retrieval metrics use the top-k retrieved results compared against ground truth relevance judgments.

Base Class

from autorag_research.evaluation.metrics import BaseRetrievalMetricConfig
from dataclasses import dataclass


@dataclass
class MyMetricConfig(BaseRetrievalMetricConfig):
    def get_metric_func(self):
        return my_metric_function