feat: Implement RAGEvaluationHarness and related classes

haystack_experimental/evaluation/__init__.py

@@ -2,6 +2,6 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-from .harness import EvalRunOverrides, EvaluationHarness
+from .harness import EvaluationHarness, EvaluationRunOverrides
 
-_all_ = ["EvaluationHarness", "EvalRunOverrides"]
+_all_ = ["EvaluationHarness", "EvaluationRunOverrides"]

haystack_experimental/evaluation/harness/__init__.py

@@ -2,6 +2,6 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-from .evalution_harness import EvalRunOverrides, EvaluationHarness
+from .evalution_harness import EvaluationHarness, EvaluationRunOverrides
 
-_all_ = ["EvaluationHarness", "EvalRunOverrides"]
+_all_ = ["EvaluationHarness", "EvaluationRunOverrides"]

haystack_experimental/evaluation/harness/evalution_harness.py

@@ -8,11 +8,10 @@
 
 from haystack import Pipeline
 from haystack.core.serialization import DeserializationCallbacks
-from haystack.evaluation.eval_run_result import BaseEvaluationRunResult
 
 
 @dataclass
-class EvalRunOverrides:
+class EvaluationRunOverrides:
     """
     Overrides for an evaluation run.
 
@@ -32,7 +31,7 @@ class EvalRunOverrides:
 
 
 EvalRunInputT = TypeVar("EvalRunInputT")
-EvalRunOutputT = TypeVar("EvalRunOutputT", bound=BaseEvaluationRunResult)
+EvalRunOutputT = TypeVar("EvalRunOutputT")
 EvalRunOverridesT = TypeVar("EvalRunOverridesT")
 
 
@@ -43,9 +42,7 @@ class EvaluationHarness(ABC, Generic[EvalRunInputT, EvalRunOverridesT, EvalRunOu
 
     @staticmethod
     def _override_pipeline(pipeline: Pipeline, parameter_overrides: Optional[Dict[str, Any]]) -> Pipeline:
-        def component_pre_init_callback(
-            name: str, cls: Type, init_params: Dict[str, Any]
-        ):  # pylint: disable=unused-argument
+        def component_pre_init_callback(name: str, cls: Type, init_params: Dict[str, Any]):  # pylint: disable=unused-argument
             assert parameter_overrides is not None
             overrides = parameter_overrides.get(name)
             if overrides:
@@ -70,7 +67,11 @@ def validate_overrides():
 
     @abstractmethod
     def run(
-        self, inputs: EvalRunInputT, *, overrides: Optional[EvalRunOverridesT] = None, run_name: Optional[str] = None
+        self,
+        inputs: EvalRunInputT,
+        *,
+        overrides: Optional[EvalRunOverridesT] = None,
+        run_name: Optional[str] = None,
     ) -> EvalRunOutputT:
         """
         Launch a evaluation run.

haystack_experimental/evaluation/harness/rag/__init__.py

@@ -0,0 +1,23 @@
+# SPDX-FileCopyrightText: 2022-present deepset GmbH <[email protected]>
+#
+# SPDX-License-Identifier: Apache-2.0
+
+from .harness import RAGEvaluationHarness
+from .parameters import (
+    RAGEvaluationInput,
+    RAGEvaluationMetric,
+    RAGEvaluationOutput,
+    RAGEvaluationOverrides,
+    RAGExpectedComponent,
+    RAGExpectedComponentMetadata,
+)
+
+_all_ = [
+    "RAGEvaluationHarness",
+    "RAGExpectedComponent",
+    "RAGExpectedComponentMetadata",
+    "RAGEvaluationMetric",
+    "RAGEvaluationOutput",
+    "RAGEvaluationOverrides",
+    "RAGEvaluationInput",
+]

haystack_experimental/evaluation/harness/rag/evaluation_pipeline.py

@@ -0,0 +1,49 @@
+# SPDX-FileCopyrightText: 2022-present deepset GmbH <[email protected]>
+#
+# SPDX-License-Identifier: Apache-2.0
+
+from functools import partial
+from typing import Set
+
+from haystack import Pipeline
+from haystack.components.evaluators import (
+    DocumentMAPEvaluator,
+    DocumentMRREvaluator,
+    DocumentRecallEvaluator,
+    FaithfulnessEvaluator,
+    SASEvaluator,
+)
+from haystack.components.evaluators.document_recall import RecallMode
+
+from .parameters import RAGEvaluationMetric
+
+
+def default_rag_evaluation_pipeline(
+    metrics: Set[RAGEvaluationMetric],
+) -> Pipeline:
+    """
+    Builds the default evaluation pipeline for RAG.
+
+    :param metrics:
+        The set of metrics to include in the pipeline.
+    :returns:
+        The evaluation pipeline.
+    """
+    pipeline = Pipeline()
+
+    metric_ctors = {
+        RAGEvaluationMetric.DOCUMENT_MAP: DocumentMAPEvaluator,
+        RAGEvaluationMetric.DOCUMENT_MRR: DocumentMRREvaluator,
+        RAGEvaluationMetric.DOCUMENT_RECALL_SINGLE_HIT: partial(DocumentRecallEvaluator, mode=RecallMode.SINGLE_HIT),
+        RAGEvaluationMetric.DOCUMENT_RECALL_MULTI_HIT: partial(DocumentRecallEvaluator, mode=RecallMode.MULTI_HIT),
+        RAGEvaluationMetric.SEMANTIC_ANSWER_SIMILARITY: partial(
+            SASEvaluator, model="sentence-transformers/all-MiniLM-L6-v2"
+        ),
+        RAGEvaluationMetric.ANSWER_FAITHFULNESS: FaithfulnessEvaluator,
+    }
+
+    for metric in metrics:
+        ctor = metric_ctors[metric]
+        pipeline.add_component(metric.value, ctor())
+
+    return pipeline