def _parse_pdp()

in src/sagemaker/model_card/evaluation_metric_parsers.py [0:0]

• 72 lines of code
• 8 McCabe index (conditional complexity)

    def _parse_pdp(self, explains: dict):
        """Parse pdp data in clarify explainability.

        Args:
            explains (dict): Explains metric data.
        """

        def format_metric_data(
            feature: dict,
            distribution_value: List,
            prediction_values: List,
            x_axis_name: Union[str, list],
            graph_type: str,
        ):
            """Create metric entry.

            Args:
                feature (dict): Entry in pdp list.
                distribution_value (List): y values in data distribution graph.
                prediction_values (List): y values in data prediction graph.
                x_axis_name (Union[str, list]): X axis name.
                graph_type (str): Type of graph.
            """
            metric_data = []
            metric_data.append(
                {
                    "name": f"data_distribution - feature={feature['feature_name']}",
                    "type": graph_type,
                    "value": distribution_value,
                    "x_axis_name": x_axis_name,
                    "y_axis_name": "Data distribution",
                }
            )
            for idx, label_header in enumerate(feature["label_headers"]):
                metric_data.append(
                    {
                        "name": (
                            f"model_predictions - feature={feature['feature_name']}, "
                            f"label={label_header}"
                        ),
                        "type": graph_type,
                        "value": prediction_values[idx],
                        "x_axis_name": x_axis_name,
                        "y_axis_name": label_header,
                    }
                )
            return metric_data

        metric_groups = []
        if "pdp" not in explains:
            return metric_groups

        metric_data = []
        for feature in explains["pdp"]:
            if feature["data_type"] == "numerical":
                distribution_value = list(
                    zip(feature["feature_values"], feature["data_distribution"])
                )
                prediction_values = [
                    list(zip(feature["feature_values"], feature["model_predictions"][i]))
                    for i, _ in enumerate(feature["label_headers"])
                ]
                x_axis_name = feature["feature_name"]
                metric_data.extend(
                    format_metric_data(
                        feature=feature,
                        distribution_value=distribution_value,
                        prediction_values=prediction_values,
                        x_axis_name=x_axis_name,
                        graph_type="linear_graph",
                    )
                )
            if feature["data_type"] == "categorical":
                distribution_value = feature["data_distribution"]
                prediction_values = [
                    feature["model_predictions"][i] for i, _ in enumerate(feature["label_headers"])
                ]
                x_axis_name = feature["feature_values"]
                metric_data.extend(
                    format_metric_data(
                        feature=feature,
                        distribution_value=distribution_value,
                        prediction_values=prediction_values,
                        x_axis_name=x_axis_name,
                        graph_type="bar_chart",
                    )
                )

        metric_groups.append({"name": "explanations - pdp", "metric_data": metric_data})

        return metric_groups