Function calc_pred_err

greykite/common/evaluation.py:359–405

Calculates the basic error measures in `~greykite.common.evaluation.EvaluationMetricEnum` and returns them in a dictionary. Parameters ---------- y_true : `list` [`float`] or `numpy.array` Observed values. y_pred : `list` [`float`] or `numpy.array` Model pred

(y_true, y_pred)

def calc_pred_err(y_true, y_pred):
    """Calculates the basic error measures in
    `~greykite.common.evaluation.EvaluationMetricEnum`
    and returns them in a dictionary.

    Parameters
    ----------
    y_true : `list` [`float`] or `numpy.array`
        Observed values.
    y_pred : `list` [`float`] or `numpy.array`
        Model predictions.

    Returns
    -------
    error : `dict` [`str`, `float` or None]
        Dictionary mapping
        `~greykite.common.evaluation.EvaluationMetricEnum`
        metric names to their values for the given ``y_true``
        and ``y_pred``.

        The dictionary is empty is there are no finite elements
        in ``y_true``.
    """
    y_true, y_pred = valid_elements_for_evaluation(
        reference_arrays=[y_true],
        arrays=[y_pred],
        reference_array_names="y_true",
        drop_leading_only=False,
        keep_inf=False)
    # The Silverkite Multistage model has NANs at the beginning
    # when predicting on the training data.
    # We only drop the leading NANs/infs from ``y_pred``,
    # since they are not supposed to appear in the middle.
    y_pred, y_true = valid_elements_for_evaluation(
        reference_arrays=[y_pred],
        arrays=[y_true],
        reference_array_names="y_pred",
        drop_leading_only=True,
        keep_inf=True)
    error = {}

    if len(y_true) > 0:
        for enum in EvaluationMetricEnum:
            metric_name = enum.get_metric_name()
            metric_func = enum.get_metric_func()
            error.update({metric_name: metric_func(y_true, y_pred)})
    return error


@add_finite_filter_to_scorer