ham.py

def find_ancestors(tree, code):
    """
    Recursively finds ancestors of a given class (e.g., an ISCO-08 code) in a hierarchical JSON structure.

    Args:
    - tree: A dictionary representing the hierarchical structure.
    - code: A string representing the label of the class.

    Returns:
    - A list of strings, each representing an ancestor of the input class.
    """
    ancestors = []
    current = code
    while current:
        parent = tree[current]["parent"]
        if parent:
            ancestors.append(parent)
        current = parent
    return ancestors


def calculate_hierarchical_measures(true_labels, predicted_labels, tree):
    """
    Calculates hierarchical precision, recall, and F-measure in a hierarchical structure.

    Args:
    - true_labels: A list of strings representing true class labels.
    - predicted_labels: A list of strings representing predicted class labels.
    - tree: A dictionary representing the hierarchical structure.

    Returns:
    - hP: A floating point number representing hierarchical precision.
    - hR: A floating point number representing hierarchical recall.
    - hF: A floating point number representing hierarchical F-measure.
    """

    extended_true = [set(find_ancestors(tree, code) | {code}) for code in true_labels]
    extended_pred = [
        set(find_ancestors(tree, code) | {code}) for code in predicted_labels
    ]

    true_positive = sum(len(t & p) for t, p in zip(extended_true, extended_pred))
    predicted = sum(len(p) for p in extended_pred)
    actual = sum(len(t) for t in extended_true)

    hP = true_positive / predicted if predicted else 0
    hR = true_positive / actual if actual else 0
    hF = (2 * hP * hR) / (hP + hR) if (hP + hR) else 0

    return hP, hR, hF


def hierarchical_f_measure(hP, hR, beta=1.0):
    """Calculate the hierarchical F-measure."""
    if hP + hR == 0:
        return 0
    return (beta**2 + 1) * hP * hR / (beta**2 * hP + hR)