training.py 文件源码

def create_eval_metric(metric_name: AnyStr) -> mx.metric.EvalMetric:
        """
        Creates an EvalMetric given a metric names.
        """
        # output_names refers to the list of outputs this metric should use to update itself, e.g. the softmax output
        if metric_name == C.ACCURACY:
            return utils.Accuracy(ignore_label=C.PAD_ID, output_names=[C.SOFTMAX_OUTPUT_NAME])
        elif metric_name == C.PERPLEXITY:
            return mx.metric.Perplexity(ignore_label=C.PAD_ID, output_names=[C.SOFTMAX_OUTPUT_NAME])
        else:
            raise ValueError("unknown metric name")