Source code for pipeline.utils.modelutils.metrics

"""Module to compute metrics to evaluate the classification performance.
"""
import typing
import torch


def compute_efficiency_purity(
    n_true_positives: int,
    n_truths: int,
    n_positives: int,
):
    eff = n_true_positives / n_truths if n_truths != 0.0 else 1.0
    pur = n_true_positives / n_positives if n_positives != 0.0 else 1.0

    return eff, pur


def compute_classification_efficiency_purity(
    predictions: torch.Tensor,
    truths: torch.Tensor,
) -> typing.Tuple[float, float]:
    """Compute the efficiency and purity of predictions.

    Args:
        predictions: tensor of predictions indicating whether each example is
            genuine (``True``) or fake (``False``)
        truths: what the ``predictions`` should be to be exact

    Returns:
        efficiency and purity of the predictions.
    """
    n_positives = predictions.sum().cpu().numpy()
    n_truths = truths.sum().cpu().numpy()
    n_true_positives = (truths.bool() & predictions).sum().cpu().numpy()

    return compute_efficiency_purity(
        n_true_positives=n_true_positives,
        n_truths=n_truths,
        n_positives=n_positives,
    )