azure-ai-evaluation 1.0.0__py3-none-any.whl → 1.0.0b2__py3-none-any.whl

azure/ai/evaluation/_evaluate/_telemetry/__init__.py

@@ -6,7 +6,7 @@ import functools
 import inspect
 import json
 import logging
-from typing import Callable, Dict, Literal, Optional, Union, cast
+from typing import Callable, Dict, TypeVar
 
 import pandas as pd
 from promptflow._sdk.entities._flows import FlexFlow as flex_flow
@@ -16,30 +16,31 @@ from promptflow.client import PFClient
 from promptflow.core import Prompty as prompty_core
 from typing_extensions import ParamSpec
 
-from azure.ai.evaluation._model_configurations import AzureAIProject, EvaluationResult
-
 from ..._user_agent import USER_AGENT
 from .._utils import _trace_destination_from_project_scope
 
 LOGGER = logging.getLogger(__name__)
 
 P = ParamSpec("P")
+R = TypeVar("R")
 
 
-def _get_evaluator_type(evaluator: Dict[str, Callable]) -> Literal["content-safety", "built-in", "custom"]:
+def _get_evaluator_type(evaluator: Dict[str, Callable]):
     """
     Get evaluator type for telemetry.
 
     :param evaluator: The evaluator object
     :type evaluator: Dict[str, Callable]
     :return: The evaluator type. Possible values are "built-in", "custom", and "content-safety".
-    :rtype: Literal["content-safety", "built-in", "custom"]
+    :rtype: str
     """
-    module = inspect.getmodule(evaluator)
-    module_name = module.__name__ if module else ""
+    built_in = False
+    content_safety = False
 
-    built_in = module_name.startswith("azure.ai.evaluation._evaluators.")
-    content_safety = built_in and module_name.startswith("azure.ai.evaluation._evaluators._content_safety")
+    module = inspect.getmodule(evaluator)
+    built_in = module and module.__name__.startswith("azure.ai.evaluation._evaluators.")
+    if built_in:
+        content_safety = module.__name__.startswith("azure.ai.evaluation._evaluators._content_safety")
 
     if content_safety:
         return "content-safety"
@@ -97,22 +98,22 @@ def _get_evaluator_properties(evaluator, evaluator_name):
 
 
 # cspell:ignore isna
-def log_evaluate_activity(func: Callable[P, EvaluationResult]) -> Callable[P, EvaluationResult]:
+def log_evaluate_activity(func: Callable[P, R]) -> Callable[P, R]:
     """Decorator to log evaluate activity
 
     :param func: The function to be decorated
     :type func: Callable
     :returns: The decorated function
-    :rtype: Callable[P, EvaluationResult]
+    :rtype: Callable[P, R]
     """
 
     @functools.wraps(func)
-    def wrapper(*args: P.args, **kwargs: P.kwargs) -> EvaluationResult:
+    def wrapper(*args: P.args, **kwargs: P.kwargs) -> R:
         from promptflow._sdk._telemetry import ActivityType, log_activity
         from promptflow._sdk._telemetry.telemetry import get_telemetry_logger
 
-        evaluators = cast(Optional[Dict[str, Callable]], kwargs.get("evaluators", {})) or {}
-        azure_ai_project = cast(Optional[AzureAIProject], kwargs.get("azure_ai_project", None))
+        evaluators = kwargs.get("evaluators", [])
+        azure_ai_project = kwargs.get("azure_ai_project", None)
 
         pf_client = PFClient(
             config=(
@@ -123,11 +124,10 @@ def log_evaluate_activity(func: Callable[P, EvaluationResult]) -> Callable[P, Ev
             user_agent=USER_AGENT,
         )
 
-        trace_destination = pf_client._config.get_trace_destination()  # pylint: disable=protected-access
-        track_in_cloud = bool(trace_destination) if trace_destination != "none" else False
+        track_in_cloud = bool(pf_client._config.get_trace_destination())  # pylint: disable=protected-access
         evaluate_target = bool(kwargs.get("target", None))
         evaluator_config = bool(kwargs.get("evaluator_config", None))
-        custom_dimensions: Dict[str, Union[str, bool]] = {
+        custom_dimensions = {
             "track_in_cloud": track_in_cloud,
             "evaluate_target": evaluate_target,
             "evaluator_config": evaluator_config,

azure/ai/evaluation/_evaluate/_utils.py

@@ -6,23 +6,15 @@ import logging
 import os
 import re
 import tempfile
+from collections import namedtuple
 from pathlib import Path
-from typing import Any, Dict, NamedTuple, Optional, Tuple, Union
-import uuid
-import base64
+from typing import Dict
 
 import pandas as pd
-from promptflow.client import PFClient
-from promptflow.entities import Run
-
-from azure.ai.evaluation._constants import (
-    DEFAULT_EVALUATION_RESULTS_FILE_NAME,
-    DefaultOpenEncoding,
-    EvaluationRunProperties,
-    Prefixes,
-)
+
+from azure.ai.evaluation._constants import DEFAULT_EVALUATION_RESULTS_FILE_NAME, DefaultOpenEncoding, Prefixes
+from azure.ai.evaluation._evaluate._eval_run import EvalRun
 from azure.ai.evaluation._exceptions import ErrorBlame, ErrorCategory, ErrorTarget, EvaluationException
-from azure.ai.evaluation._model_configurations import AzureAIProject
 
 LOGGER = logging.getLogger(__name__)
 
@@ -31,20 +23,14 @@ AZURE_WORKSPACE_REGEX_FORMAT = (
     "(/providers/Microsoft.MachineLearningServices)?/workspaces/([^/]+)$"
 )
 
-
-class AzureMLWorkspace(NamedTuple):
-    subscription_id: str
-    resource_group_name: str
-    workspace_name: str
+AzureMLWorkspaceTriad = namedtuple("AzureMLWorkspace", ["subscription_id", "resource_group_name", "workspace_name"])
 
 
-def is_none(value) -> bool:
+def is_none(value):
     return value is None or str(value).lower() == "none"
 
 
-def extract_workspace_triad_from_trace_provider(  # pylint: disable=name-too-long
-    trace_provider: str,
-) -> AzureMLWorkspace:
+def extract_workspace_triad_from_trace_provider(trace_provider: str):  # pylint: disable=name-too-long
     match = re.match(AZURE_WORKSPACE_REGEX_FORMAT, trace_provider)
     if not match or len(match.groups()) != 5:
         raise EvaluationException(
@@ -61,7 +47,7 @@ def extract_workspace_triad_from_trace_provider(  # pylint: disable=name-too-lon
     subscription_id = match.group(1)
     resource_group_name = match.group(3)
     workspace_name = match.group(5)
-    return AzureMLWorkspace(subscription_id, resource_group_name, workspace_name)
+    return AzureMLWorkspaceTriad(subscription_id, resource_group_name, workspace_name)
 
 
 def load_jsonl(path):
@@ -69,7 +55,7 @@ def load_jsonl(path):
         return [json.loads(line) for line in f.readlines()]
 
 
-def _azure_pf_client_and_triad(trace_destination) -> Tuple[PFClient, AzureMLWorkspace]:
+def _azure_pf_client_and_triad(trace_destination):
     from promptflow.azure._cli._utils import _get_azure_pf_client
 
     ws_triad = extract_workspace_triad_from_trace_provider(trace_destination)
@@ -82,45 +68,15 @@ def _azure_pf_client_and_triad(trace_destination) -> Tuple[PFClient, AzureMLWork
     return azure_pf_client, ws_triad
 
 
-def _store_multimodal_content(messages, tmpdir: str):
-    # verify if images folder exists
-    images_folder_path = os.path.join(tmpdir, "images")
-    os.makedirs(images_folder_path, exist_ok=True)
-
-    # traverse all messages and replace base64 image data with new file name.
-    for message in messages:
-        if isinstance(message.get("content", []), list):
-            for content in message.get("content", []):
-                if content.get("type") == "image_url":
-                    image_url = content.get("image_url")
-                    if image_url and "url" in image_url and image_url["url"].startswith("data:image/jpg;base64,"):
-                        # Extract the base64 string
-                        base64image = image_url["url"].replace("data:image/jpg;base64,", "")
-
-                        # Generate a unique filename
-                        image_file_name = f"{str(uuid.uuid4())}.jpg"
-                        image_url["url"] = f"images/{image_file_name}"  # Replace the base64 URL with the file path
-
-                        # Decode the base64 string to binary image data
-                        image_data_binary = base64.b64decode(base64image)
-
-                        # Write the binary image data to the file
-                        image_file_path = os.path.join(images_folder_path, image_file_name)
-                        with open(image_file_path, "wb") as f:
-                            f.write(image_data_binary)
-
-
 def _log_metrics_and_instance_results(
-    metrics: Dict[str, Any],
-    instance_results: pd.DataFrame,
-    trace_destination: Optional[str],
-    run: Run,
-    evaluation_name: Optional[str],
-) -> Optional[str]:
-    from azure.ai.evaluation._evaluate._eval_run import EvalRun
-
+    metrics,
+    instance_results,
+    trace_destination,
+    run,
+    evaluation_name,
+) -> str:
     if trace_destination is None:
-        LOGGER.debug("Skip uploading evaluation results to AI Studio since no trace destination was provided.")
+        LOGGER.error("Unable to log traces as trace destination was not defined.")
         return None
 
     azure_pf_client, ws_triad = _azure_pf_client_and_triad(trace_destination)
@@ -138,18 +94,10 @@ def _log_metrics_and_instance_results(
         ml_client=azure_pf_client.ml_client,
         promptflow_run=run,
     ) as ev_run:
-        artifact_name = EvalRun.EVALUATION_ARTIFACT
+
+        artifact_name = EvalRun.EVALUATION_ARTIFACT if run else EvalRun.EVALUATION_ARTIFACT_DUMMY_RUN
 
         with tempfile.TemporaryDirectory() as tmpdir:
-            # storing multi_modal images if exists
-            col_name = "inputs.conversation"
-            if col_name in instance_results.columns:
-                for item in instance_results[col_name].items():
-                    value = item[1]
-                    if "messages" in value:
-                        _store_multimodal_content(value["messages"], tmpdir)
-
-            # storing artifact result
             tmp_path = os.path.join(tmpdir, artifact_name)
 
             with open(tmp_path, "w", encoding=DefaultOpenEncoding.WRITE) as f:
@@ -164,9 +112,9 @@ def _log_metrics_and_instance_results(
             if run is None:
                 ev_run.write_properties_to_run_history(
                     properties={
-                        EvaluationRunProperties.RUN_TYPE: "eval_run",
-                        EvaluationRunProperties.EVALUATION_RUN: "promptflow.BatchRun",
+                        "_azureml.evaluation_run": "azure-ai-generative-parent",
                         "_azureml.evaluate_artifacts": json.dumps([{"path": artifact_name, "type": "table"}]),
+                        "isEvaluatorRun": "true",
                     }
                 )
 
@@ -190,7 +138,7 @@ def _get_ai_studio_url(trace_destination: str, evaluation_id: str) -> str:
     return studio_url
 
 
-def _trace_destination_from_project_scope(project_scope: AzureAIProject) -> str:
+def _trace_destination_from_project_scope(project_scope: dict) -> str:
     subscription_id = project_scope["subscription_id"]
     resource_group_name = project_scope["resource_group_name"]
     workspace_name = project_scope["project_name"]
@@ -203,19 +151,17 @@ def _trace_destination_from_project_scope(project_scope: AzureAIProject) -> str:
     return trace_destination
 
 
-def _write_output(path: Union[str, os.PathLike], data_dict: Any) -> None:
+def _write_output(path, data_dict):
     p = Path(path)
-    if p.is_dir():
+    if os.path.isdir(path):
         p = p / DEFAULT_EVALUATION_RESULTS_FILE_NAME
 
     with open(p, "w", encoding=DefaultOpenEncoding.WRITE) as f:
         json.dump(data_dict, f)
 
-    print(f'Evaluation results saved to "{p.resolve()}".\n')
-
 
 def _apply_column_mapping(
-    source_df: pd.DataFrame, mapping_config: Optional[Dict[str, str]], inplace: bool = False
+    source_df: pd.DataFrame, mapping_config: Dict[str, str], inplace: bool = False
 ) -> pd.DataFrame:
     """
     Apply column mapping to source_df based on mapping_config.
@@ -265,7 +211,7 @@ def _apply_column_mapping(
     return result_df
 
 
-def _has_aggregator(evaluator: object) -> bool:
+def _has_aggregator(evaluator):
     return hasattr(evaluator, "__aggregate__")
 
 
@@ -288,11 +234,11 @@ def get_int_env_var(env_var_name: str, default_value: int) -> int:
         return default_value
 
 
-def set_event_loop_policy() -> None:
+def set_event_loop_policy():
     import asyncio
     import platform
 
     if platform.system().lower() == "windows":
         # Reference: https://stackoverflow.com/questions/45600579/asyncio-event-loop-is-closed-when-getting-loop
         # On Windows seems to be a problem with EventLoopPolicy, use this snippet to work around it
-        asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())  # type: ignore[attr-defined]
+        asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())

azure/ai/evaluation/_evaluators/_bleu/_bleu.py

@@ -26,29 +26,30 @@ class _AsyncBleuScoreEvaluator:
 
 class BleuScoreEvaluator:
     """
-    Calculate the BLEU score for a given response and ground truth.
+    Evaluator that computes the BLEU Score between two strings.
 
     BLEU (Bilingual Evaluation Understudy) score is commonly used in natural language processing (NLP) and machine
-    translation. It is widely used in text summarization and text generation use cases.
+    translation. It is widely used in text summarization and text generation use cases. It evaluates how closely the
+    generated text matches the reference text. The BLEU score ranges from 0 to 1, with higher scores indicating
+    better quality.
 
-    Use the BLEU score when you want to evaluate the similarity between the generated text and reference text,
-    especially in tasks such as machine translation or text summarization, where n-gram overlap is a significant
-    indicator of quality.
+    **Usage**
 
-    The BLEU score ranges from 0 to 1, with higher scores indicating better quality.
+    .. code-block:: python
 
-    .. admonition:: Example:
+        eval_fn = BleuScoreEvaluator()
+        result = eval_fn(
+            response="Tokyo is the capital of Japan.",
+            ground_truth="The capital of Japan is Tokyo.")
 
-        .. literalinclude:: ../samples/evaluation_samples_evaluate.py
-            :start-after: [START bleu_score_evaluator]
-            :end-before: [END bleu_score_evaluator]
-            :language: python
-            :dedent: 8
-            :caption: Initialize and call an BleuScoreEvaluator.
-    """
+    **Output format**
+
+    .. code-block:: python
 
-    id = "azureml://registries/azureml/models/Bleu-Score-Evaluator/versions/3"
-    """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
+        {
+            "bleu_score": 0.22
+        }
+    """
 
     def __init__(self):
         self._async_evaluator = _AsyncBleuScoreEvaluator()
@@ -62,7 +63,7 @@ class BleuScoreEvaluator:
         :keyword ground_truth: The ground truth to be compared against.
         :paramtype ground_truth: str
         :return: The BLEU score.
-        :rtype: Dict[str, float]
+        :rtype: dict
         """
         return async_run_allowing_running_loop(
             self._async_evaluator, response=response, ground_truth=ground_truth, **kwargs

azure/ai/evaluation/_evaluators/{_retrieval → _chat}/__init__.py

@@ -2,8 +2,8 @@
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # ---------------------------------------------------------
 
-from ._retrieval import RetrievalEvaluator
+from ._chat import ChatEvaluator
 
 __all__ = [
-    "RetrievalEvaluator",
+    "ChatEvaluator",
 ]