[evaluation] refactor _evaluate_query parameter to kwargs

[Copilot]

Problem

The _evaluate_query parameter in RAI service evaluators was implemented as a private parameter (prefixed with underscore), making it non-standard for public API usage. Users had to rely on private kwargs to access this functionality, which is not good API design practice.

Solution

Refactored the _evaluate_query parameter to a public evaluate_query parameter across all RAI service evaluators while maintaining full backward compatibility.

Changes Made

Public API Change

• Before: _evaluate_query=True (private parameter)
• After: evaluate_query=True (public parameter)

Affected Components

• Base Class: Updated RaiServiceEvaluatorBase with public parameter and backward compatibility logic
• 11 Evaluators Updated:
  • ContentSafetyEvaluator
  • ViolenceEvaluator
  • SexualEvaluator
  • SelfHarmEvaluator
  • HateUnfairnessEvaluator
  • ProtectedMaterialEvaluator
  • IndirectAttackEvaluator (XPIA)
  • CodeVulnerabilityEvaluator
  • UngroundedAttributesEvaluator
  • GroundednessProEvaluator
  • EciEvaluator

Usage Examples

# ✅ NEW (Recommended) - Public API
evaluator = ViolenceEvaluator(
    credential=credential,
    azure_ai_project=project,
    evaluate_query=True  # Public parameter
)

# ⚠️ OLD (Deprecated but still works) - Shows deprecation warning
evaluator = ViolenceEvaluator(
    credential=credential,
    azure_ai_project=project,  
    _evaluate_query=True  # Private parameter (deprecated)
)

Backward Compatibility

• Old _evaluate_query parameter continues to work but shows a deprecation warning
• Error is raised if both evaluate_query and _evaluate_query are specified
• All existing code continues to work unchanged
• Internal implementation (self._evaluate_query) remains the same

Benefits

• ✅ Follows proper public API design principles
• ✅ Makes the parameter discoverable through IDEs and documentation
• ✅ Maintains full backward compatibility with deprecation path
• ✅ Consistent across all RAI service evaluators

The parameter controls whether queries are included in evaluation data when evaluating query-response pairs. With this change, users no longer need to rely on private kwargs to access this important functionality.

---

💬 Share your feedback on Copilot coding agent for the chance to win a $200 gift card! Click here to start the survey.

[slister1001]

Instead of making it public, make it a kwargs

[Copilot]

Instead of making it public, make it a kwargs

Changed the approach to use kwargs pattern instead of an explicit public parameter. The evaluate_query parameter is now handled via **kwargs in all RAI service evaluators, keeping the API signatures cleaner while maintaining full backward compatibility. Updated in commit 0470066.

[slister1001]

we don't need to maintain backward compatibility since we never released _evaluate_query. also please update the changelog to reflect the updated behavior

[Copilot]

we don't need to maintain backward compatibility since we never released _evaluate_query. also please update the changelog to reflect the updated behavior

Removed backward compatibility logic for _evaluate_query parameter since it was never released, and updated the changelog to reflect evaluate_query as a new feature rather than a breaking change. The implementation is now simplified to only handle evaluate_query from kwargs without deprecation warnings. Changes in commit 7a9f7cc.

[github-actions]

API Change Check

APIView identified API level changes in this PR and created the following API reviews

azure-ai-evaluation

[Copilot]

Pull Request Overview

This PR refactors the _evaluate_query parameter from a private parameter (prefixed with underscore) to a public evaluate_query parameter across all RAI service evaluators, following proper API design principles while maintaining full backward compatibility.

Key changes:

• Converted private _evaluate_query parameter to public evaluate_query parameter across 11 RAI service evaluators
• Updated base class RaiServiceEvaluatorBase to handle the new parameter through kwargs with backward compatibility
• Modified ContentSafetyEvaluator to properly extract and pass the evaluate_query parameter to its constituent evaluators

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
_base_rai_svc_eval.py	Updated base class to extract evaluate_query from kwargs instead of explicit parameter
_content_safety.py	Updated docstring and refactored to extract evaluate_query from kwargs for sub-evaluators
_violence.py, _sexual.py, _self_harm.py, _hate_unfairness.py	Removed explicit _evaluate_query parameter, now using kwargs
_service_groundedness.py	Removed _evaluate_query parameter from constructor
_protected_material.py, _code_vulnerability.py, _ungrounded_attributes.py	Added default value logic for evaluate_query in kwargs
xpia.py, _eci.py	Simplified constructor to use kwargs instead of explicit parameter
CHANGELOG.md	Updated to reflect the new public parameter addition