-
Add Trajectory Accuracy LLM-as-a-Judge Metric (#2529) (#2539)
frostbyte_neo 发布于
2025-07-29 09:03:49 +00:00 -
Add trajectory accuracy metric for ReAct-style agent evaluation
-
Complete TrajectoryAccuracy integration: add docs and finalize implementation
- Added comprehensive documentation following Opik patterns
- Updated metrics overview with TrajectoryAccuracy entry
- Created dedicated trajectory_accuracy.mdx documentation page
- Added Python SDK documentation files
- Removed unnecessary standalone demo files
- All tests pass and implementation follows established patterns
- Ready for review and merge
-
Refactor: Standardize TrajectoryAccuracy metric and clean up example
-
Refactor trajectory accuracy tests to eliminate private attribute access │
│ │
│ - Replace private method calls with public API equivalents: │
│ * _parse_evaluation_response() -> parser.parse_evaluation_response() │
│ * _format_trajectory_steps() -> templates.create_evaluation_prompt() │
│ - Use dependency injection for mock model via constructor instead of setting private │
│ _model attribute │
│ - Test behavior through public interfaces rather than accessing private state │
│ - Add proper mock specification with base_model.OpikBaseModel │
│ - Update error handling test to match current MetricComputationError behavior │
│ │
│ Addresses code review feedback about private/protected attribute access violations. │
│ All tests pass and follow proper encapsulation principles. -
Resolve merge conflicts in evaluation metrics and add trajectory metrics
下载附件
-