Results

The Results dataclass aggregates the metrics computed on a surrogate’s predictions relative to the targets (for example, accuracy, recall, regret), storing per-round metrics and summary statistics. The metrics computed depend on whether the task is classification or regression; see alf_core.utils.metrics for more details and examples. It is the primary return value of a completed ALF experiment and is used to evaluate and compare experiment runs.

class alf_core.dataclasses.results.Results(targets, predictions, problem_type)

Bases: object

Computes metrics based on the predictions and targets.

targets

A numpy array of ground truth target values.

predictions

A Predictions object containing model predictions.

problem_type

Type of problem determining which metrics are computed.

compute_metrics()

Compute evaluation metrics based on predictions and targets.

For regression, routes to the variance-aware regression registry. For classification (binary or multiclass), routes to the classification metric registry using the probability array stored in predictions.means.

Raises:

ValueError – If the problem type is unrecognized.

Return type:

dict[str, Union[float, int, number]]

Returns:

A dictionary of metric names to their computed values.

predictions: Predictions

problem_type: ProblemType

targets: ndarray