def from_row()

in src/fmeval/eval_algorithms/util.py [0:0]
18 lines of code
7 McCabe index (conditional complexity)

    def from_row(row: Dict[str, Union[str, float, int]], score_names: List[str]) -> "EvalOutputRecord":
        """
        Returns an instance of EvalOutputRecord, created from a Ray Dataset row (represented as a dict).

        Example input:
            row = {
                "model_input": "input",
                "model_output": "output",
                "column_that_wont_be_included": "hello",
                "rouge": 0.42,
                "bert": 0.162
            }

        Corresponding output:
            EvalOutputRecord(
                scores=[
                    EvalScore(name="rouge", value=0.42),
                    EvalScore(name="bert", value=0.162)
                ],
                dataset_columns={
                    "model_input": "input",
                    "model_output": "output"
                }
            )

        Note how "column_that_wont_be_included" is not included in the produced EvalOutputRecord.
        This is because only columns in constants.COLUMN_NAMES are considered to be valid columns
        in the saved output file generated by `save_dataset`. The reason why it's even possible
        for a column name that doesn't belong to constants.COLUMN_NAMES to appear in `row` is that
        the Ray Dataset that `row` belongs to can contain columns used to store intermediate computations.
        For example, ClassificationAccuracy generates a column named CLASSIFIED_MODEL_OUTPUT_COLUMN_NAME
        that is used to compute CLASSIFICATION_ACCURACY_SCORE, which is one of the score columns.

        :param row: a Ray Dataset row represented as a dict
        :param score_names: column names included in the Ray Dataset that `row`
            is a sample of that correspond to evaluation algorithm scores
        :returns: an instance of EvalOutputRecord corresponding to `row`
        """
        dataset_columns = {}
        scores = []
        for column_name, value in row.items():
            if column_name not in score_names:  # pragma: no branch
                if column_name in DATASET_COLUMNS:  # pragma: no branch
                    dataset_columns[column_name] = value
            else:
                assert isinstance(value, float) or isinstance(value, int) or value is None  # to satisfy Mypy
                if value is None:
                    assert row.get(DatasetColumns.ERROR.value.name, None)
                    scores.append(EvalScore(name=column_name, error=row.get(DatasetColumns.ERROR.value.name)))
                else:
                    scores.append(EvalScore(name=column_name, value=value))

        return EvalOutputRecord(
            scores=scores,
            dataset_columns=dataset_columns,
        )