def compute_precision_recall()

in evalbench/scorers/recallmatcher.py [0:0]

• 34 lines of code
• 11 McCabe index (conditional complexity)

    def compute_precision_recall(self, golden_results, generated_results):
        """
        Calculates precision and recall for two sets of results, removing duplicates and Nones.

        Args:
            golden_results: A list of the correct results.
            generated_results: A list of the results generated by the model.

        Returns:
            A dictionary containing the following:
                1. precision
                2. recall
                3. orig_golden_size: no. of expected results before removing duplicates
                4. orig_generated_size: no. of generated results before removing duplicates
                5. dedup_golden_size: no. of expected results after removing duplicates
                6. dedup_generated_size: no. of generated results after removing duplicates
        """

        # Filter out None values (assuming they shouldn't be considered)
        golden_results = (
            [x for x in golden_results if x is not None] if golden_results else []
        )
        generated_results = (
            [x for x in generated_results if x is not None] if generated_results else []
        )

        orig_golden_size = len(golden_results)
        orig_generated_size = len(generated_results)

        golden_results_set = convert_to_set(golden_results)
        generated_results_set = convert_to_set(generated_results)

        dedup_golden_size = len(golden_results_set)
        dedup_generated_size = len(generated_results_set)

        common_results_set = golden_results_set.intersection(generated_results_set)
        correct_predictions = len(common_results_set)

        if golden_results == generated_results:
            precision = 1
            recall = 1
        else:
            # Calculate precision and recall
            recall = (
                correct_predictions / len(generated_results) if generated_results else 0
            )
            precision = (
                correct_predictions / len(golden_results) if golden_results else 0
            )

        full_result = {
            "precision": precision,
            "recall": recall,
            "orig_golden_size": orig_golden_size,
            "orig_generated_size": orig_generated_size,
            "dedup_golden_size": dedup_golden_size,
            "dedup_generated_size": dedup_generated_size,
        }
        return full_result