|
| 1 | +"""Balanced accuracy evaluator — custom evaluator with non-trivial score aggregation. |
| 2 | +
|
| 3 | +Balanced accuracy = mean of per-class recall rates. |
| 4 | +
|
| 5 | +Per-datapoint scores encode class weights: |
| 6 | + - correct prediction: score = 1 / (num_classes * class_count_for_expected) |
| 7 | + - wrong prediction: score = 0 |
| 8 | +
|
| 9 | +Then reduce_scores sums the scores, which yields: |
| 10 | + sum = Σ_k (correct_k / (K * n_k)) = (1/K) Σ_k (correct_k / n_k) = balanced_accuracy |
| 11 | +""" |
| 12 | + |
| 13 | +from uipath.eval.evaluators.base_evaluator import ( |
| 14 | + BaseEvaluationCriteria, |
| 15 | + BaseEvaluatorJustification, |
| 16 | +) |
| 17 | +from uipath.eval.evaluators.output_evaluator import ( |
| 18 | + BaseOutputEvaluator, |
| 19 | + OutputEvaluatorConfig, |
| 20 | +) |
| 21 | +from uipath.eval.models import ( |
| 22 | + AgentExecution, |
| 23 | + EvaluationResult, |
| 24 | + NumericEvaluationResult, |
| 25 | +) |
| 26 | +from uipath.eval.models.models import ( |
| 27 | + EvaluationResultDto, |
| 28 | + UiPathEvaluationError, |
| 29 | + UiPathEvaluationErrorCategory, |
| 30 | +) |
| 31 | + |
| 32 | + |
| 33 | +class BalancedAccuracyEvaluationCriteria(BaseEvaluationCriteria): |
| 34 | + """Per-datapoint criteria: which class this sample should belong to.""" |
| 35 | + |
| 36 | + expected_class: str |
| 37 | + |
| 38 | + |
| 39 | +class BalancedAccuracyEvaluatorConfig( |
| 40 | + OutputEvaluatorConfig[BalancedAccuracyEvaluationCriteria] |
| 41 | +): |
| 42 | + """Evaluator config with class list and per-class sample counts.""" |
| 43 | + |
| 44 | + name: str = "BalancedAccuracyEvaluator" |
| 45 | + classes: list[str] |
| 46 | + class_counts: dict[str, int] |
| 47 | + |
| 48 | + |
| 49 | +class BalancedAccuracyJustification(BaseEvaluatorJustification): |
| 50 | + """Details about how this datapoint was scored.""" |
| 51 | + |
| 52 | + predicted_class: str |
| 53 | + expected_class: str |
| 54 | + weight: float |
| 55 | + is_match: bool |
| 56 | + |
| 57 | + |
| 58 | +class BalancedAccuracyEvaluator( |
| 59 | + BaseOutputEvaluator[ |
| 60 | + BalancedAccuracyEvaluationCriteria, |
| 61 | + BalancedAccuracyEvaluatorConfig, |
| 62 | + BalancedAccuracyJustification, |
| 63 | + ] |
| 64 | +): |
| 65 | + """Balanced accuracy: mean of per-class recall rates. |
| 66 | +
|
| 67 | + Uses weighted per-datapoint scores so that reduce_scores = sum() |
| 68 | + gives the correct balanced accuracy. |
| 69 | + """ |
| 70 | + |
| 71 | + @classmethod |
| 72 | + def get_evaluator_id(cls) -> str: |
| 73 | + """Get the evaluator id.""" |
| 74 | + return "custom-balanced-accuracy" |
| 75 | + |
| 76 | + @staticmethod |
| 77 | + def reduce_scores(results: list[EvaluationResultDto]) -> float: |
| 78 | + """Sum of pre-weighted scores = balanced accuracy.""" |
| 79 | + return sum(r.score for r in results) |
| 80 | + |
| 81 | + async def evaluate( |
| 82 | + self, |
| 83 | + agent_execution: AgentExecution, |
| 84 | + evaluation_criteria: BalancedAccuracyEvaluationCriteria, |
| 85 | + ) -> EvaluationResult: |
| 86 | + predicted_class = str(self._get_actual_output(agent_execution)).lower() |
| 87 | + expected_class = evaluation_criteria.expected_class.lower() |
| 88 | + classes = [c.lower() for c in self.evaluator_config.classes] |
| 89 | + class_counts = { |
| 90 | + k.lower(): v for k, v in self.evaluator_config.class_counts.items() |
| 91 | + } |
| 92 | + |
| 93 | + if expected_class not in classes: |
| 94 | + raise UiPathEvaluationError( |
| 95 | + code="INVALID_EXPECTED_CLASS", |
| 96 | + title="Expected class not in configured classes", |
| 97 | + detail=f"Expected class '{expected_class}' is not in the configured classes: {classes}", |
| 98 | + category=UiPathEvaluationErrorCategory.USER, |
| 99 | + ) |
| 100 | + |
| 101 | + if predicted_class not in classes: |
| 102 | + raise UiPathEvaluationError( |
| 103 | + code="INVALID_PREDICTED_CLASS", |
| 104 | + title="Predicted class not in configured classes", |
| 105 | + detail=f"Predicted class '{predicted_class}' is not in the configured classes: {classes}", |
| 106 | + category=UiPathEvaluationErrorCategory.USER, |
| 107 | + ) |
| 108 | + |
| 109 | + num_classes = len(classes) |
| 110 | + n_k = class_counts.get(expected_class) |
| 111 | + if n_k is None or n_k <= 0: |
| 112 | + raise UiPathEvaluationError( |
| 113 | + code="INVALID_CLASS_COUNT", |
| 114 | + title="Missing or invalid class count", |
| 115 | + detail=f"class_counts must include a positive count for '{expected_class}'", |
| 116 | + category=UiPathEvaluationErrorCategory.USER, |
| 117 | + ) |
| 118 | + |
| 119 | + weight = 1.0 / (num_classes * n_k) |
| 120 | + is_match = predicted_class == expected_class |
| 121 | + score = weight if is_match else 0.0 |
| 122 | + |
| 123 | + justification = self.validate_justification( |
| 124 | + { |
| 125 | + "expected": expected_class, |
| 126 | + "actual": predicted_class, |
| 127 | + "predicted_class": predicted_class, |
| 128 | + "expected_class": expected_class, |
| 129 | + "weight": weight, |
| 130 | + "is_match": is_match, |
| 131 | + } |
| 132 | + ) |
| 133 | + return NumericEvaluationResult(score=score, details=justification) |
0 commit comments