eval.ragas_bridge

chunktuner.eval.ragas_bridge

RAGAS Faithfulness + Answer Relevancy (optional chunktuner[ragas] + compatible stack).

RagasBridge 

RagasBridge(llm_client=None)

Thin adapter around RAGAS faithfulness and answer_relevancy (optional extra).

Source code in src/chunktuner/eval/ragas_bridge.py 
18
19
def __init__(self, llm_client: object | None = None):
    self.llm_client = llm_client

compute 

compute(questions, contexts, answers, ground_truths)

Return averaged metrics, or sentinel Nones if RAGAS cannot run.

Source code in src/chunktuner/eval/ragas_bridge.py 
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
def compute(
    self,
    questions: list[str],
    contexts: list[list[str]],
    answers: list[str],
    ground_truths: list[str],
) -> dict[str, float | None]:
    """Return averaged metrics, or sentinel Nones if RAGAS cannot run."""
    try:
        from datasets import Dataset
        from ragas import evaluate
        from ragas.metrics import answer_relevancy, faithfulness
    except ImportError as exc:  # pragma: no cover
        logger.warning("RAGAS not installed (%s). Install: uv add 'chunktuner[ragas]'", exc)
        return dict(_UNAVAILABLE)
    except Exception as exc:  # pragma: no cover
        logger.warning("RAGAS import failed (%s); skipping generation metrics", exc)
        return dict(_UNAVAILABLE)

    try:
        rows = []
        for q, ctx, ans, gt in zip(questions, contexts, answers, ground_truths, strict=True):
            rows.append(
                {
                    "user_input": q,
                    "retrieved_contexts": ctx,
                    "response": ans,
                    "reference": gt,
                }
            )
        ds = Dataset.from_list(rows)
        result = evaluate(ds, metrics=[faithfulness, answer_relevancy])
        df = result.to_pandas()
        out: dict[str, float | None] = dict(_UNAVAILABLE)
        if "faithfulness" in df.columns:
            out["faithfulness"] = float(df["faithfulness"].mean())
        if "answer_relevancy" in df.columns:
            out["answer_relevancy"] = float(df["answer_relevancy"].mean())
        return out
    except Exception as exc:  # pragma: no cover
        logger.warning("RAGAS computation failed: %s", exc)
        return dict(_UNAVAILABLE)