azure-ai-evaluation 1.0.1__py3-none-any.whl → 1.2.0__py3-none-any.whl

azure/ai/evaluation/_evaluators/_content_safety/_hate_unfairness.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -71,6 +72,7 @@ class HateUnfairnessEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.HATE_FAIRNESS,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_content_safety/_self_harm.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -65,6 +66,7 @@ class SelfHarmEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.SELF_HARM,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_content_safety/_sexual.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -67,6 +68,7 @@ class SexualEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.SEXUAL,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_content_safety/_violence.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -67,6 +68,7 @@ class ViolenceEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.VIOLENCE,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_f1_score/_f1_score.py

@@ -3,45 +3,44 @@
 # ---------------------------------------------------------
 
 from collections import Counter
-from typing import List
+from typing import List, Dict
+from typing_extensions import overload, override
 
-from promptflow._utils.async_utils import async_run_allowing_running_loop
+from azure.ai.evaluation._evaluators._common import EvaluatorBase
 
-from azure.ai.evaluation._exceptions import ErrorBlame, ErrorCategory, ErrorTarget, EvaluationException
 
+class F1ScoreEvaluator(EvaluatorBase):
+    """
+    Calculates the F1 score for a given response and ground truth or a multi-turn conversation.
 
-class _AsyncF1ScoreEvaluator:
-    def __init__(self):
-        pass
+    F1 Scores range from 0 to 1, with 1 being the best possible score.
 
-    async def __call__(self, *, response: str, ground_truth: str, **kwargs):
-        """
-        Evaluate F1 score.
+    The F1-score computes the ratio of the number of shared words between the model generation and
+    the ground truth. Ratio is computed over the individual words in the generated response against those in the ground
+    truth answer. The number of shared words between the generation and the truth is the basis of the F1 score:
+    precision is the ratio of the number of shared words to the total number of words in the generation, and recall
+    is the ratio of the number of shared words to the total number of words in the ground truth.
 
-        :keyword response: The response to be evaluated.
-        :paramtype response: str
-        :keyword ground_truth: The ground truth to be evaluated.
-        :paramtype ground_truth: str
-        :return: The F1 score.
-        :rtype: Dict[str, float]
-        """
-        # Validate inputs
-        if not (response and response.strip() and response != "None") or not (
-            ground_truth and ground_truth.strip() and ground_truth != "None"
-        ):
-            msg = "Both 'response' and 'ground_truth' must be non-empty strings."
-            raise EvaluationException(
-                message=msg,
-                internal_message=msg,
-                error_category=ErrorCategory.MISSING_FIELD,
-                error_blame=ErrorBlame.USER_ERROR,
-                error_target=ErrorTarget.F1_EVALUATOR,
-            )
+    Use the F1 score when you want a single comprehensive metric that combines both recall and precision in your
+    model's responses. It provides a balanced evaluation of your model's performance in terms of capturing accurate
+    information in the response.
 
-        # Run f1 score computation.
-        f1_result = self._compute_f1_score(response=response, ground_truth=ground_truth)
 
-        return {"f1_score": f1_result}
+    .. admonition:: Example:
+
+        .. literalinclude:: ../samples/evaluation_samples_evaluate.py
+            :start-after: [START f1_score_evaluator]
+            :end-before: [END f1_score_evaluator]
+            :language: python
+            :dedent: 8
+            :caption: Initialize and call an F1ScoreEvaluator.
+    """
+
+    id = "azureml://registries/azureml/models/F1Score-Evaluator/versions/3"
+    """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
+
+    def __init__(self):
+        super().__init__()
 
     @classmethod
     def _compute_f1_score(cls, response: str, ground_truth: str) -> float:
@@ -103,41 +102,24 @@ class _AsyncF1ScoreEvaluator:
 
         return f1
 
+    @override
+    async def _do_eval(self, eval_input: Dict) -> Dict[str, float]:
+        """Produce an f1 score evaluation result.
 
-class F1ScoreEvaluator:
-    """
-    Calculates the F1 score for a given response and ground truth or a multi-turn conversation.
-
-    F1 Scores range from 0 to 1, with 1 being the best possible score.
-
-    The F1-score computes the ratio of the number of shared words between the model generation and
-    the ground truth. Ratio is computed over the individual words in the generated response against those in the ground
-    truth answer. The number of shared words between the generation and the truth is the basis of the F1 score:
-    precision is the ratio of the number of shared words to the total number of words in the generation, and recall
-    is the ratio of the number of shared words to the total number of words in the ground truth.
-
-    Use the F1 score when you want a single comprehensive metric that combines both recall and precision in your
-    model's responses. It provides a balanced evaluation of your model's performance in terms of capturing accurate
-    information in the response.
-
-
-    .. admonition:: Example:
-
-        .. literalinclude:: ../samples/evaluation_samples_evaluate.py
-            :start-after: [START f1_score_evaluator]
-            :end-before: [END f1_score_evaluator]
-            :language: python
-            :dedent: 8
-            :caption: Initialize and call an F1ScoreEvaluator.
-    """
-
-    id = "azureml://registries/azureml/models/F1Score-Evaluator/versions/3"
-    """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
+        :param eval_input: The input to the evaluation function.
+        :type eval_input: Dict
+        :return: The evaluation result.
+        :rtype: Dict
+        """
+        ground_truth = eval_input["ground_truth"]
+        response = eval_input["response"]
+        # Run f1 score computation.
+        f1_result = self._compute_f1_score(response=response, ground_truth=ground_truth)
 
-    def __init__(self):
-        self._async_evaluator = _AsyncF1ScoreEvaluator()
+        return {"f1_score": f1_result}
 
-    def __call__(self, *, response: str, ground_truth: str, **kwargs):
+    @overload  # type: ignore
+    def __call__(self, *, response: str, ground_truth: str) -> Dict[str, float]:
         """
         Evaluate F1 score.
 
@@ -149,9 +131,20 @@ class F1ScoreEvaluator:
         :rtype: Dict[str, float]
         """
 
-        return async_run_allowing_running_loop(
-            self._async_evaluator, response=response, ground_truth=ground_truth, **kwargs
-        )
+    @override
+    def __call__(  # pylint: disable=docstring-missing-param
+        self,
+        *args,
+        **kwargs,
+    ):
+        """
+        Evaluate F1 score.
 
-    def _to_async(self):
-        return self._async_evaluator
+        :keyword response: The response to be evaluated.
+        :paramtype response: str
+        :keyword ground_truth: The ground truth to be evaluated.
+        :paramtype ground_truth: str
+        :return: The F1 score.
+        :rtype: Dict[str, float]
+        """
+        return super().__call__(*args, **kwargs)

azure/ai/evaluation/_evaluators/_gleu/_gleu.py

@@ -1,28 +1,16 @@
 # ---------------------------------------------------------
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # ---------------------------------------------------------
+from typing import Dict
 from nltk.translate.gleu_score import sentence_gleu
-from promptflow._utils.async_utils import async_run_allowing_running_loop
+from typing_extensions import overload, override
 
 from azure.ai.evaluation._common.utils import nltk_tokenize
 
+from azure.ai.evaluation._evaluators._common import EvaluatorBase
 
-class _AsyncGleuScoreEvaluator:
-    def __init__(self):
-        pass
-
-    async def __call__(self, *, ground_truth: str, response: str, **kwargs):
-        reference_tokens = nltk_tokenize(ground_truth)
-        hypothesis_tokens = nltk_tokenize(response)
-
-        score = sentence_gleu([reference_tokens], hypothesis_tokens)
-
-        return {
-            "gleu_score": score,
-        }
 
-
-class GleuScoreEvaluator:
+class GleuScoreEvaluator(EvaluatorBase):
     """
     Calculates the GLEU (Google-BLEU) score between a response and the ground truth.
 
@@ -47,10 +35,32 @@ class GleuScoreEvaluator:
     id = "azureml://registries/azureml/models/Gleu-Score-Evaluator/versions/3"
     """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
 
+    @override
     def __init__(self):
-        self._async_evaluator = _AsyncGleuScoreEvaluator()
+        super().__init__()
+
+    @override
+    async def _do_eval(self, eval_input: Dict) -> Dict[str, float]:
+        """Produce a glue score evaluation result.
+
+        :param eval_input: The input to the evaluation function.
+        :type eval_input: Dict
+        :return: The evaluation result.
+        :rtype: Dict
+        """
+        ground_truth = eval_input["ground_truth"]
+        response = eval_input["response"]
+        reference_tokens = nltk_tokenize(ground_truth)
+        hypothesis_tokens = nltk_tokenize(response)
 
-    def __call__(self, *, ground_truth: str, response: str, **kwargs):
+        score = sentence_gleu([reference_tokens], hypothesis_tokens)
+
+        return {
+            "gleu_score": score,
+        }
+
+    @overload  # type: ignore
+    def __call__(self, *, ground_truth: str, response: str):
         """
         Evaluate the GLEU score between the response and the ground truth.
 
@@ -61,9 +71,21 @@ class GleuScoreEvaluator:
         :return: The GLEU score.
         :rtype: Dict[str, float]
         """
-        return async_run_allowing_running_loop(
-            self._async_evaluator, ground_truth=ground_truth, response=response, **kwargs
-        )
 
-    def _to_async(self):
-        return self._async_evaluator
+    @override
+    def __call__(  # pylint: disable=docstring-missing-param
+        self,
+        *args,
+        **kwargs,
+    ):
+        """
+        Evaluate the GLEU score between the response and the ground truth.
+
+        :keyword response: The response to be evaluated.
+        :paramtype response: str
+        :keyword ground_truth: The ground truth to be compared against.
+        :paramtype ground_truth: str
+        :return: The GLEU score.
+        :rtype: Dict[str, float]
+        """
+        return super().__call__(*args, **kwargs)

azure/ai/evaluation/_evaluators/_meteor/_meteor.py

@@ -1,38 +1,16 @@
 # ---------------------------------------------------------
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # ---------------------------------------------------------
+from typing import Dict
+
 from nltk.translate.meteor_score import meteor_score
-from promptflow._utils.async_utils import async_run_allowing_running_loop
+from typing_extensions import overload, override
 
 from azure.ai.evaluation._common.utils import nltk_tokenize, ensure_nltk_data_downloaded
+from azure.ai.evaluation._evaluators._common import EvaluatorBase
 
 
-class _AsyncMeteorScoreEvaluator:
-    def __init__(self, alpha: float = 0.9, beta: float = 3.0, gamma: float = 0.5):
-        self._alpha = alpha
-        self._beta = beta
-        self._gamma = gamma
-
-        ensure_nltk_data_downloaded()
-
-    async def __call__(self, *, ground_truth: str, response: str, **kwargs):
-        reference_tokens = nltk_tokenize(ground_truth)
-        hypothesis_tokens = nltk_tokenize(response)
-
-        score = meteor_score(
-            [reference_tokens],
-            hypothesis_tokens,
-            alpha=self._alpha,
-            beta=self._beta,
-            gamma=self._gamma,
-        )
-
-        return {
-            "meteor_score": score,
-        }
-
-
-class MeteorScoreEvaluator:
+class MeteorScoreEvaluator(EvaluatorBase):
     """
     Calculates the METEOR score for a given response and ground truth.
 
@@ -68,10 +46,41 @@ class MeteorScoreEvaluator:
     id = "azureml://registries/azureml/models/Meteor-Score-Evaluator/versions/3"
     """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
 
+    @override
     def __init__(self, alpha: float = 0.9, beta: float = 3.0, gamma: float = 0.5):
-        self._async_evaluator = _AsyncMeteorScoreEvaluator(alpha=alpha, beta=beta, gamma=gamma)
+        self._alpha = alpha
+        self._beta = beta
+        self._gamma = gamma
+        ensure_nltk_data_downloaded()
+        super().__init__()
 
-    def __call__(self, *, ground_truth: str, response: str, **kwargs):
+    @override
+    async def _do_eval(self, eval_input: Dict) -> Dict[str, float]:
+        """Produce a meteor score evaluation result.
+
+        :param eval_input: The input to the evaluation function.
+        :type eval_input: Dict
+        :return: The evaluation result.
+        :rtype: Dict
+        """
+        ground_truth = eval_input["ground_truth"]
+        response = eval_input["response"]
+        reference_tokens = nltk_tokenize(ground_truth)
+        hypothesis_tokens = nltk_tokenize(response)
+        score = meteor_score(
+            [reference_tokens],
+            hypothesis_tokens,
+            alpha=self._alpha,
+            beta=self._beta,
+            gamma=self._gamma,
+        )
+
+        return {
+            "meteor_score": score,
+        }
+
+    @overload  # type: ignore
+    def __call__(self, *, ground_truth: str, response: str) -> Dict[str, float]:
         """
         Evaluate the METEOR score between the response and the ground truth.
 
@@ -82,9 +91,21 @@ class MeteorScoreEvaluator:
         :return: The METEOR score.
         :rtype: Dict[str, float]
         """
-        return async_run_allowing_running_loop(
-            self._async_evaluator, ground_truth=ground_truth, response=response, **kwargs
-        )
 
-    def _to_async(self):
-        return self._async_evaluator
+    @override
+    def __call__(  # pylint: disable=docstring-missing-param
+        self,
+        *args,
+        **kwargs,
+    ):
+        """
+        Evaluate the METEOR score between the response and the ground truth.
+
+        :keyword response: The response to be evaluated.
+        :paramtype response: str
+        :keyword ground_truth: The ground truth to be compared against.
+        :paramtype ground_truth: str
+        :return: The METEOR score.
+        :rtype: Dict[str, float]
+        """
+        return super().__call__(*args, **kwargs)

azure/ai/evaluation/_evaluators/_qa/_qa.py

@@ -2,10 +2,11 @@
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # ---------------------------------------------------------
 
-from concurrent.futures import as_completed
-from typing import Callable, Dict, List, Union
+from typing import Union
 
-from promptflow.tracing import ThreadPoolExecutorWithContext as ThreadPoolExecutor
+from typing_extensions import overload, override
+
+from azure.ai.evaluation._evaluators._common import MultiEvaluatorBase
 
 from .._coherence import CoherenceEvaluator
 from .._f1_score import F1ScoreEvaluator
@@ -15,7 +16,7 @@ from .._relevance import RelevanceEvaluator
 from .._similarity import SimilarityEvaluator
 
 
-class QAEvaluator:
+class QAEvaluator(MultiEvaluatorBase[Union[str, float]]):
     """
     Initialize a question-answer evaluator configured for a specific Azure OpenAI model.
 
@@ -46,9 +47,7 @@ class QAEvaluator:
     """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
 
     def __init__(self, model_config, **kwargs):
-        self._parallel = kwargs.pop("_parallel", False)
-
-        self._evaluators: List[Union[Callable[..., Dict[str, Union[str, float]]], Callable[..., Dict[str, float]]]] = [
+        evaluators = [
             GroundednessEvaluator(model_config),
             RelevanceEvaluator(model_config),
             CoherenceEvaluator(model_config),
@@ -56,8 +55,31 @@ class QAEvaluator:
             SimilarityEvaluator(model_config),
             F1ScoreEvaluator(),
         ]
+        super().__init__(evaluators=evaluators, **kwargs)
+
+    @overload  # type: ignore
+    def __call__(self, *, query: str, response: str, context: str, ground_truth: str):
+        """
+        Evaluates question-answering scenario.
+
+        :keyword query: The query to be evaluated.
+        :paramtype query: str
+        :keyword response: The response to be evaluated.
+        :paramtype response: str
+        :keyword context: The context to be evaluated.
+        :paramtype context: str
+        :keyword ground_truth: The ground truth to be evaluated.
+        :paramtype ground_truth: str
+        :return: The scores for QA scenario.
+        :rtype: Dict[str, Union[str, float]]
+        """
 
-    def __call__(self, *, query: str, response: str, context: str, ground_truth: str, **kwargs):
+    @override
+    def __call__(  # pylint: disable=docstring-missing-param
+        self,
+        *args,
+        **kwargs,
+    ):
         """
         Evaluates question-answering scenario.
 
@@ -72,22 +94,5 @@ class QAEvaluator:
         :return: The scores for QA scenario.
         :rtype: Dict[str, Union[str, float]]
         """
-        results: Dict[str, Union[str, float]] = {}
-        if self._parallel:
-            with ThreadPoolExecutor() as executor:
-                futures = {
-                    executor.submit(
-                        evaluator, query=query, response=response, context=context, ground_truth=ground_truth, **kwargs
-                    ): evaluator
-                    for evaluator in self._evaluators
-                }
-
-                # Collect results as they complete
-                for future in as_completed(futures):
-                    results.update(future.result())
-        else:
-            for evaluator in self._evaluators:
-                result = evaluator(query=query, response=response, context=context, ground_truth=ground_truth, **kwargs)
-                results.update(result)
-
-        return results
+
+        return super().__call__(*args, **kwargs)

azure/ai/evaluation/_evaluators/_rouge/_rouge.py

@@ -3,9 +3,11 @@
 # ---------------------------------------------------------
 from enum import Enum
 
-from promptflow._utils.async_utils import async_run_allowing_running_loop
+from typing import Dict
+from typing_extensions import overload, override
 
 from azure.ai.evaluation._vendor.rouge_score import rouge_scorer
+from azure.ai.evaluation._evaluators._common import EvaluatorBase
 
 
 class RougeType(Enum):
@@ -32,21 +34,7 @@ class RougeType(Enum):
     """Overlap of L-grams (L consecutive words) between generated and reference text."""
 
 
-class _AsyncRougeScoreEvaluator:
-    def __init__(self, rouge_type: RougeType):
-        self._rouge_type = rouge_type
-
-    async def __call__(self, *, ground_truth: str, response: str, **kwargs):
-        scorer = rouge_scorer.RougeScorer(rouge_types=[self._rouge_type.value])
-        metrics = scorer.score(ground_truth, response)[self._rouge_type.value]
-        return {
-            "rouge_precision": metrics.precision,
-            "rouge_recall": metrics.recall,
-            "rouge_f1_score": metrics.fmeasure,
-        }
-
-
-class RougeScoreEvaluator:
+class RougeScoreEvaluator(EvaluatorBase):
     """
     Calculates the ROUGE score for a given response and ground truth.
 
@@ -76,10 +64,32 @@ class RougeScoreEvaluator:
     id = "azureml://registries/azureml/models/Rouge-Score-Evaluator/versions/3"
     """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
 
+    @override
     def __init__(self, rouge_type: RougeType):
-        self._async_evaluator = _AsyncRougeScoreEvaluator(rouge_type)
+        self._rouge_type = rouge_type
+        super().__init__()
+
+    @override
+    async def _do_eval(self, eval_input: Dict) -> Dict[str, float]:
+        """Produce a rouge score evaluation result.
 
-    def __call__(self, *, ground_truth: str, response: str, **kwargs):
+        :param eval_input: The input to the evaluation function.
+        :type eval_input: Dict
+        :return: The evaluation result.
+        :rtype: Dict
+        """
+        ground_truth = eval_input["ground_truth"]
+        response = eval_input["response"]
+        scorer = rouge_scorer.RougeScorer(rouge_types=[self._rouge_type.value])
+        metrics = scorer.score(ground_truth, response)[self._rouge_type.value]
+        return {
+            "rouge_precision": metrics.precision,
+            "rouge_recall": metrics.recall,
+            "rouge_f1_score": metrics.fmeasure,
+        }
+
+    @overload  # type: ignore
+    def __call__(self, *, ground_truth: str, response: str) -> Dict[str, float]:
         """
         Evaluate the ROUGE score between the response and the ground truth.
 
@@ -90,9 +100,20 @@ class RougeScoreEvaluator:
         :return: The ROUGE score.
         :rtype: Dict[str, float]
         """
-        return async_run_allowing_running_loop(
-            self._async_evaluator, ground_truth=ground_truth, response=response, **kwargs
-        )
 
-    def _to_async(self):
-        return self._async_evaluator
+    @override
+    def __call__(  # pylint: disable=docstring-missing-param
+        self,
+        *args,
+        **kwargs,
+    ):
+        """
+        Evaluate route score.
+        :keyword response: The response to be evaluated.
+        :paramtype response: str
+        :keyword ground_truth: The ground truth to be compared against.
+        :paramtype ground_truth: str
+        :return: The ROUGE score.
+        :rtype: Dict[str, float]
+        """
+        return super().__call__(*args, **kwargs)