Bugfix Summarization Score (explodinggradients#1164)

Hi! 

I was experimenting with Summarization Score and found an issue that
when there are no questions generated by the LLM, it throws division by
zero error due to 0 generated answers.

## Notable Changes
- Added an assertion to keyphrases, questions and answers generation
functions.
- Changed the llm response variable name `answer` to `response` to
prevent possible confusion with score related `answer`.

Please let me know if there is something I can add. 

Cheers!

---------

Co-authored-by: jjmachan <[email protected]>

docs/howtos/customisations/run_config.ipynb

@@ -53,15 +53,16 @@
     "\n",
     "# load the dataset\n",
     "from datasets import load_dataset\n",
+    "\n",
     "amnesty_qa = load_dataset(\"explodinggradients/amnesty_qa\", \"english_v2\")\n",
     "\n",
     "# configure RunConfig\n",
     "from ragas.run_config import RunConfig\n",
     "\n",
     "_ = evaluate(\n",
-    "    dataset=amnesty_qa[\"eval\"], \n",
+    "    dataset=amnesty_qa[\"eval\"],\n",
     "    metrics=[faithfulness],\n",
-    "    run_config=RunConfig(max_workers=64), # increasing max_workers from default 16\n",
+    "    run_config=RunConfig(max_workers=64),  # increasing max_workers from default 16\n",
     ")"
    ]
   },
@@ -94,9 +95,9 @@
    ],
    "source": [
     "_ = evaluate(\n",
-    "    dataset=amnesty_qa[\"eval\"], \n",
+    "    dataset=amnesty_qa[\"eval\"],\n",
     "    metrics=[faithfulness],\n",
-    "    run_config=RunConfig(max_workers=2), # increasing max_workers from default 16\n",
+    "    run_config=RunConfig(max_workers=2),  # increasing max_workers from default 16\n",
     ")"
    ]
   },

src/ragas/metrics/_summarization.py

@@ -186,7 +186,7 @@ def _compute_score(self, scores) -> float:
         """Returns average score of the different scores."""
         return sum(scores) / len(scores)
 
-    def _compute_qa_score(self, answers: t.List) -> float:
+    def _compute_qa_score(self, answers: t.List[str]) -> float:
         """Returns a score between 0 and 1 reflecting the fraction of
         correct answers, ie with a value 'yes'
         """
@@ -209,10 +209,15 @@ async def _extract_keyphrases(self, text: str, callbacks: Callbacks) -> t.List[s
             callbacks=callbacks,
         )
         result_text = result.generations[0][0].text
-        answer = await _output_parser_keyphrase_extraction.aparse(
+        response = await _output_parser_keyphrase_extraction.aparse(
             result_text, p_value, self.llm, self.max_retries
         )
-        return answer.keyphrases if answer else []
+
+        if not response or not response.keyphrases:
+            logging.error("No keyphrases generated, unable to calculate the score.")
+            return []
+
+        return response.keyphrases
 
     async def _get_questions(
         self, text: str, keyphrases: list[str], callbacks: Callbacks
@@ -225,13 +230,15 @@ async def _get_questions(
         )
 
         result_text = result.generations[0][0].text
-        answer = await _output_parser_question_generation.aparse(
+        response = await _output_parser_question_generation.aparse(
             result_text, p_value, self.llm, self.max_retries
         )
-        if answer is None:
+
+        if not response or not response.questions:
+            logging.error("No questions generated, unable to calculate the score.")
             return []
 
-        return answer.questions
+        return response.questions
 
     async def _get_answers(
         self, questions: t.List[str], summary: str, callbacks: Callbacks
@@ -244,13 +251,15 @@ async def _get_answers(
         )
 
         result_text = result.generations[0][0].text
-        answer = await _output_parser_answer_generation.aparse(
+        response = await _output_parser_answer_generation.aparse(
             result_text, p_value, self.llm, self.max_retries
         )
-        if answer is None:
+
+        if not response or not response.answers:
+            logger.error("No answers generated, unable to calculate the score.")
             return []
 
-        return answer.answers
+        return response.answers
 
 
 def adapt(self, language: str, cache_dir: str | None = None) -> None:

src/ragas/metrics/base.py

@@ -63,11 +63,13 @@ def get_required_columns(
 class Metric(ABC):
     @property
     @abstractmethod
-    def name(self) -> str: ...
+    def name(self) -> str:
+        ...
 
     @property
     @abstractmethod
-    def evaluation_mode(self) -> EvaluationMode: ...
+    def evaluation_mode(self) -> EvaluationMode:
+        ...
 
     @abstractmethod
     def init(self, run_config: RunConfig):
@@ -130,7 +132,8 @@ async def ascore(
         return score
 
     @abstractmethod
-    async def _ascore(self, row: t.Dict, callbacks: Callbacks) -> float: ...
+    async def _ascore(self, row: t.Dict, callbacks: Callbacks) -> float:
+        ...
 
 
 @dataclass

src/ragas/testset/prompts.py

@@ -509,5 +509,5 @@ class EvolutionElimination(BaseModel):
     question_rewrite_prompt,
     context_scoring_prompt,
     filter_question_prompt,
-    evolution_elimination_prompt
-]
+    evolution_elimination_prompt,
+]

tests/unit/test_analytics.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
+
 import typing as t
+
 import pytest
 
 
@@ -130,7 +132,7 @@ def test_testset_generation_tracking(monkeypatch):
 
 
 def test_was_completed(monkeypatch):
-    from ragas._analytics import track_was_completed, IsCompleteEvent
+    from ragas._analytics import IsCompleteEvent, track_was_completed
 
     event_properties_list: t.List[IsCompleteEvent] = []
 

tests/unit/test_executor_in_jupyter.ipynb

@@ -33,7 +33,7 @@
     "\n",
     "exec = Executor(raise_exceptions=True)\n",
     "for i in range(10):\n",
-    "    exec.submit(sleep, i/10)\n",
+    "    exec.submit(sleep, i / 10)\n",
     "\n",
     "assert exec.results(), \"didn't get anything from results\""
    ]
@@ -140,16 +140,18 @@
    "source": [
     "from ragas.metrics.base import Metric, EvaluationMode\n",
     "\n",
+    "\n",
     "class FakeMetric(Metric):\n",
     "    name = \"fake_metric\"\n",
     "    evaluation_mode = EvaluationMode.qa\n",
     "\n",
     "    def init(self):\n",
     "        pass\n",
     "\n",
-    "    async def _ascore(self, row, callbacks)->float:\n",
+    "    async def _ascore(self, row, callbacks) -> float:\n",
     "        return 0\n",
     "\n",
+    "\n",
     "fm = FakeMetric()"
    ]
   },

tests/unit/test_run_config.py

@@ -1,36 +1,42 @@
-import sys, importlib
-from packaging.version import parse as parse_version
+import importlib
+import sys
 from platform import python_version
+
 import pytest
-from numpy.random import default_rng, Generator
+from numpy.random import Generator, default_rng
+from packaging.version import parse as parse_version
 
 from ragas.run_config import RunConfig
 
 if parse_version(python_version()) < parse_version("3.10"):
-    from typing import NewType, Callable
-    RandomComparison = NewType("RandomComparison", Callable[[Generator, Generator], bool])
+    from typing import Callable, NewType
+
+    RandomComparison = NewType(
+        "RandomComparison", Callable[[Generator, Generator], bool]
+    )
 elif parse_version(python_version()) >= parse_version("3.10"):
-    from typing import TypeAlias, Callable
+    from typing import Callable, TypeAlias
+
     RandomComparison: TypeAlias = Callable[[Generator, Generator], bool]
 
+
 @pytest.fixture(scope="function")
 def compare_rng() -> Callable[[Generator, Generator], bool]:
-    """Pytest fixture wrapper to check :py:cls:`numpy.random.Generator` object equivalence.
+    """Pytest fixture wrapper to check :py:cls:`numpy.random.Generator` object equivalence."""
 
-    """
-    def _compare_rng(rng_0:Generator, rng_1:Generator) -> bool:
+    def _compare_rng(rng_0: Generator, rng_1: Generator) -> bool:
         """Compare two :py:cls:`numpy.random.Generator`object.
-        
+
         Args:
             rng_0 (numpy.random.Generator) : The first generator to compare with.
             rng_1 (numpy.random.Generator) : The second generator to compare with.
 
         Returns:
             bool: Whether the two generators are at the same state.
-                
+
         """
         return rng_0.random() == rng_1.random()
-    
+
     return _compare_rng
 
 
@@ -39,9 +45,11 @@ def _compare_rng(rng_0:Generator, rng_1:Generator) -> bool:
     (
         [42, True],
         [None, False],
-    )
+    ),
 )
-def test_random_num_generator(seed, compare_rng:RandomComparison, expected_equivalence):
+def test_random_num_generator(
+    seed, compare_rng: RandomComparison, expected_equivalence
+):
     """Check :py:mod:`numpy.random` functionality and seed behaviour control."""
     rc = RunConfig(seed=seed)
 
@@ -53,7 +61,7 @@ def test_random_num_generator(seed, compare_rng:RandomComparison, expected_equiv
     assert compare_rng(rc.rng, rng) == expected_equivalence
 
     # Check generation consistency
-    importlib.reload(sys.modules['numpy.random'])
+    importlib.reload(sys.modules["numpy.random"])
     new_rc = RunConfig(seed=seed)
     new_rng = default_rng(seed=seed)
 
@@ -63,22 +71,14 @@ def test_random_num_generator(seed, compare_rng:RandomComparison, expected_equiv
 
     # Check equivalence
     if expected_equivalence:
-        assert all(
-             list(
-                  map(
-                       compare_rng,
-                       [rc.rng, new_rc.rng],
-                       [new_rng, rng]
-                    )
-                )
-            )
+        assert all(list(map(compare_rng, [rc.rng, new_rc.rng], [new_rng, rng])))
     else:
         assert all(
-             list(
-                  map(
-                       lambda x, y:not compare_rng(x, y),
-                       [rc.rng, new_rc.rng],
-                       [new_rng, rng]
-                    )
+            list(
+                map(
+                    lambda x, y: not compare_rng(x, y),
+                    [rc.rng, new_rc.rng],
+                    [new_rng, rng],
                 )
             )
+        )