​
Overview

The BaseMetric class provides an abstract base for implementing individual metrics in prompt evaluation. It defines a common interface for computing metrics based on dataset items and predictions.

​
Methods

compute
method

Abstract method to compute the metric.

Parameters:

  • dataset_item (DatasetItem): The dataset item to evaluate. It includes inputs, outputs, and metadata as attributes.
  • pred (Union[str, Dict[str, Any]]): The prediction to evaluate.

Returns: MetricResult: An object containing the computed score and any intermediate values.

Note: The implementation of this method should handle the comparison between pred and dataset_item["outputs"], potentially using the information in dataset_item["inputs"] and dataset_item["metadata"] to inform the computation. It should return a MetricResult object that encapsulates the result of the metric calculation.

__call__
method

Unified method to compute the metric, handling both sync and async implementations.

Parameters:

  • dataset_item (DatasetItem): The dataset item to evaluate.
  • pred (Union[str, Dict[str, Any]]): The prediction to evaluate.

Returns: MetricResult: An object containing the score and intermediate values.

​
Usage

To use the BaseMetric class, create a subclass and implement the compute method:

from ape.common.metrics import BaseMetric

class MyMetric(BaseMetric):
    def compute(self, dataset_item: DatasetItem, pred: Union[str, Dict[str, Any]]) -> MetricResult:
        # Implement your metric computation here
        return MetricResult(score=0.5)

Was this page helpful?

GeneratorJsonMatchMetric