def hook()

in quant/utils/visualization.py [0:0]

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

    def hook(self, split: str, metrics: Dict[str, Metric],
             epoch: int, global_step: int, log_interval: int = 10,
             values_dict: Optional[Dict[str, float]] = None, **kwargs: Any) -> None:
        """
        Provide a training / test loop-compatible hook for logging evaluation metrics.

        Args:
            split: The split to visualize, e.g. train or test
            metrics: Dictionary mapping metric names to Metric objects
            epoch: Training epoch
            global_step: Unique incrementing integer across all epochs indicating the step
            log_interval: frequency for logging metrics
            values_dict: Dictionary mapping names to values
                for other non-metric values to log
        """
        if values_dict is None:
            values_dict = {}

        if split != 'train':
            for name, metric in metrics.items():
                name = name.replace(' ', '_')
                self.writer.add_scalar(f'{name}/{split}', metric.compute(), epoch)

            for name, val in values_dict.items():
                name = name.replace(' ', '_')
                self.writer.add_scalar(f'{name}/{split}', val, epoch)

        elif global_step % log_interval == 0:
            for name, metric in metrics.items():
                name = name.replace(' ', '_')
                self.writer.add_scalar(f'{name}/{split}', metric.compute(), global_step)

            for name, val in values_dict.items():
                name = name.replace(' ', '_')
                self.writer.add_scalar(f'{name}/{split}', val, global_step)