feat: Add context evaluation to eval framework [COG-1366] (#586)

<!-- .github/pull_request_template.md -->

## Description
<!-- Provide a clear description of the changes in this PR -->

## DCO Affirmation
I affirm that all code in every commit of this pull request conforms to
the terms of the Topoteretes Developer Certificate of Origin


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

- **New Features**
- Introduced a class-based retrieval mechanism to enhance answer
generation with improved context extraction and completion.
- Added a new evaluation metric for contextual relevancy and an option
to enable context evaluation during the evaluation process.

- **Refactor**
- Transitioned from a function-based answer resolver to a more modular
retriever approach to improve extensibility.

- **Tests**
- Updated tests to align with the new answer generation and evaluation
process.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: lxobr <122801072+lxobr@users.noreply.github.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: Daniel Molnar <soobrosa@gmail.com>
Co-authored-by: Boris <boris@topoteretes.com>

cognee/eval_framework/answer_generation/answer_generation_executor.py

@@ -1,21 +1,17 @@
-import cognee
-from typing import List, Dict, Callable, Awaitable
-from cognee.api.v1.search import SearchType
+from typing import List, Dict
+from cognee.modules.retrieval.completion_retriever import CompletionRetriever
+from cognee.modules.retrieval.graph_completion_retriever import GraphCompletionRetriever
+from cognee.modules.retrieval.graph_summary_completion_retriever import (
+    GraphSummaryCompletionRetriever,
+)
 
-question_answering_engine_options: Dict[str, Callable[[str, str], Awaitable[List[str]]]] = {
-    "cognee_graph_completion": lambda query, system_prompt_path: cognee.search(
-        query_type=SearchType.GRAPH_COMPLETION,
-        query_text=query,
-        system_prompt_path=system_prompt_path,
-    ),
-    "cognee_completion": lambda query, system_prompt_path: cognee.search(
-        query_type=SearchType.COMPLETION, query_text=query, system_prompt_path=system_prompt_path
-    ),
-    "graph_summary_completion": lambda query, system_prompt_path: cognee.search(
-        query_type=SearchType.GRAPH_SUMMARY_COMPLETION,
-        query_text=query,
-        system_prompt_path=system_prompt_path,
-    ),
+from cognee.modules.retrieval.base_retriever import BaseRetriever
+
+
+retriever_options: Dict[str, BaseRetriever] = {
+    "cognee_graph_completion": GraphCompletionRetriever,
+    "cognee_completion": CompletionRetriever,
+    "graph_summary_completion": GraphSummaryCompletionRetriever,
 }
 
 
@@ -23,20 +19,22 @@ class AnswerGeneratorExecutor:
     async def question_answering_non_parallel(
         self,
         questions: List[Dict[str, str]],
-        answer_resolver: Callable[[str], Awaitable[List[str]]],
+        retriever: BaseRetriever,
     ) -> List[Dict[str, str]]:
         answers = []
         for instance in questions:
             query_text = instance["question"]
             correct_answer = instance["answer"]
 
-            search_results = await answer_resolver(query_text)
+            retrieval_context = await retriever.get_context(query_text)
+            search_results = await retriever.get_completion(query_text, retrieval_context)
 
             answers.append(
                 {
                     "question": query_text,
                     "answer": search_results[0],
                     "golden_answer": correct_answer,
+                    "retrieval_context": retrieval_context,
                 }
             )
 

cognee/eval_framework/answer_generation/run_question_answering_module.py

@@ -3,7 +3,7 @@ import json
 from typing import List
 from cognee.eval_framework.answer_generation.answer_generation_executor import (
     AnswerGeneratorExecutor,
-    question_answering_engine_options,
+    retriever_options,
 )
 from cognee.infrastructure.files.storage import LocalStorage
 from cognee.infrastructure.databases.relational.get_relational_engine import (
@@ -48,9 +48,7 @@ async def run_question_answering(
         answer_generator = AnswerGeneratorExecutor()
         answers = await answer_generator.question_answering_non_parallel(
             questions=questions,
-            answer_resolver=lambda query: question_answering_engine_options[params["qa_engine"]](
-                query, system_prompt
-            ),
+            retriever=retriever_options[params["qa_engine"]](system_prompt_path=system_prompt),
         )
         with open(params["answers_path"], "w", encoding="utf-8") as f:
             json.dump(answers, f, ensure_ascii=False, indent=4)

cognee/eval_framework/eval_config.py

@@ -18,6 +18,7 @@ class EvalConfig(BaseSettings):
 
     # Evaluation params
     evaluating_answers: bool = True
+    evaluating_contexts: bool = True
     evaluation_engine: str = "DeepEval"  # Options: 'DeepEval' (uses deepeval_model), 'DirectLLM' (uses default llm from .env)
     evaluation_metrics: List[str] = [
         "correctness",
@@ -51,6 +52,7 @@ class EvalConfig(BaseSettings):
             "answering_questions": self.answering_questions,
             "qa_engine": self.qa_engine,
             "evaluating_answers": self.evaluating_answers,
+            "evaluating_contexts": self.evaluating_contexts,  # Controls whether context evaluation should be performed
             "evaluation_engine": self.evaluation_engine,
             "evaluation_metrics": self.evaluation_metrics,
             "calculate_metrics": self.calculate_metrics,

cognee/eval_framework/evaluation/deep_eval_adapter.py

@@ -5,6 +5,7 @@ from cognee.eval_framework.evaluation.base_eval_adapter import BaseEvalAdapter
 from cognee.eval_framework.evaluation.metrics.exact_match import ExactMatchMetric
 from cognee.eval_framework.evaluation.metrics.f1 import F1ScoreMetric
 from typing import Any, Dict, List
+from deepeval.metrics import ContextualRelevancyMetric
 
 
 class DeepEvalAdapter(BaseEvalAdapter):
@@ -13,6 +14,7 @@ class DeepEvalAdapter(BaseEvalAdapter):
             "correctness": self.g_eval_correctness(),
             "EM": ExactMatchMetric(),
             "f1": F1ScoreMetric(),
+            "contextual_relevancy": ContextualRelevancyMetric(),
         }
 
     async def evaluate_answers(
@@ -29,6 +31,7 @@ class DeepEvalAdapter(BaseEvalAdapter):
                 input=answer["question"],
                 actual_output=answer["answer"],
                 expected_output=answer["golden_answer"],
+                retrieval_context=[answer["retrieval_context"]],
             )
             metric_results = {}
             for metric in evaluator_metrics:

cognee/eval_framework/evaluation/evaluation_executor.py

@@ -3,7 +3,11 @@ from cognee.eval_framework.evaluation.evaluator_adapters import EvaluatorAdapter
 
 
 class EvaluationExecutor:
-    def __init__(self, evaluator_engine: Union[str, EvaluatorAdapter, Any] = "DeepEval") -> None:
+    def __init__(
+        self,
+        evaluator_engine: Union[str, EvaluatorAdapter, Any] = "DeepEval",
+        evaluate_contexts: bool = False,
+    ) -> None:
         if isinstance(evaluator_engine, str):
             try:
                 adapter_enum = EvaluatorAdapter(evaluator_engine)
@@ -14,7 +18,10 @@ class EvaluationExecutor:
             self.eval_adapter = evaluator_engine.adapter_class()
         else:
             self.eval_adapter = evaluator_engine
+        self.evaluate_contexts = evaluate_contexts
 
     async def execute(self, answers: List[Dict[str, str]], evaluator_metrics: Any) -> Any:
+        if self.evaluate_contexts:
+            evaluator_metrics.append("contextual_relevancy")
         metrics = await self.eval_adapter.evaluate_answers(answers, evaluator_metrics)
         return metrics

cognee/eval_framework/evaluation/run_evaluation_module.py

@@ -42,7 +42,10 @@ async def execute_evaluation(params: dict) -> None:
         raise ValueError(f"Error decoding JSON from {params['answers_path']}: {e}")
 
     logging.info(f"Loaded {len(answers)} answers from {params['answers_path']}")
-    evaluator = EvaluationExecutor(evaluator_engine=params["evaluation_engine"])
+    evaluator = EvaluationExecutor(
+        evaluator_engine=params["evaluation_engine"],
+        evaluate_contexts=params["evaluating_contexts"],
+    )
     metrics = await evaluator.execute(
         answers=answers, evaluator_metrics=params["evaluation_metrics"]
     )

cognee/tests/unit/eval_framework/answer_generation_test.py

@@ -11,14 +11,18 @@ async def test_answer_generation():
     limit = 1
     corpus_list, qa_pairs = DummyAdapter().load_corpus(limit=limit)
 
-    mock_answer_resolver = AsyncMock()
-    mock_answer_resolver.side_effect = lambda query: ["mock_answer"]
+    mock_retriever = AsyncMock()
+    mock_retriever.get_context = AsyncMock(return_value="Mocked retrieval context")
+    mock_retriever.get_completion = AsyncMock(return_value=["Mocked answer"])
 
     answer_generator = AnswerGeneratorExecutor()
     answers = await answer_generator.question_answering_non_parallel(
-        questions=qa_pairs, answer_resolver=mock_answer_resolver
+        questions=qa_pairs,
+        retriever=mock_retriever,
     )
 
+    mock_retriever.get_context.assert_any_await(qa_pairs[0]["question"])
+
     assert len(answers) == len(qa_pairs)
     assert answers[0]["question"] == qa_pairs[0]["question"], (
         "AnswerGeneratorExecutor is passing the question incorrectly"
@@ -26,6 +30,6 @@ async def test_answer_generation():
     assert answers[0]["golden_answer"] == qa_pairs[0]["answer"], (
         "AnswerGeneratorExecutor is passing the golden answer incorrectly"
     )
-    assert answers[0]["answer"] == "mock_answer", (
+    assert answers[0]["answer"] == "Mocked answer", (
         "AnswerGeneratorExecutor is passing the generated answer incorrectly"
     )