src/evaluation/__init__.py

"""
RAG Evaluation Framework

Comprehensive evaluation system for RAG applications including:
- Retrieval quality metrics (Precision@K, Recall@K, MRR, NDCG)
- Generation quality metrics (BLEU, ROUGE, BERTScore, Faithfulness)
- System performance metrics (Latency, Throughput, Error rates)
- User experience metrics (Satisfaction, Task completion)
"""

"""Lazy evaluation exports to avoid heavy imports at test collection time.

Some evaluation modules (BLEU/ROUGE/BERTScore) import large packages like
`torch` which slow down pytest collection. Expose lightweight accessors that
perform imports only when evaluation functions are actually used.
"""

from .core import BenchmarkResults, EvaluationMetrics, EvaluationResult  # noqa: E402


def _lazy_import(name: str):
    module = __import__(f"src.evaluation.metrics.{name}", fromlist=["*"])
    return module


def get_generation_metrics():
    m = _lazy_import("generation_metrics")
    return (
        m.calculate_bert_score,
        m.calculate_bleu_score,
        m.calculate_faithfulness_score,
        m.calculate_rouge_scores,
    )


def get_retrieval_metrics():
    m = _lazy_import("retrieval_metrics")
    return m.mean_reciprocal_rank, m.ndcg_at_k, m.precision_at_k, m.recall_at_k


def get_system_metrics():
    m = _lazy_import("system_metrics")
    return m.ErrorTracker, m.LatencyTracker, m.ThroughputTracker


def get_user_metrics():
    m = _lazy_import("user_metrics")
    return m.CitationAccuracyTracker, m.TaskCompletionTracker, m.UserSatisfactionTracker


def get_runner():
    m = _lazy_import("runner")
    return m.EvaluationRunner


__all__ = [
    "EvaluationMetrics",
    "EvaluationResult",
    "BenchmarkResults",
    "get_generation_metrics",
    "get_retrieval_metrics",
    "get_system_metrics",
    "get_user_metrics",
    "get_runner",
]