abydos.phones._phones.cmp_features()

Complexity

Conditions

15

Size

Total Lines	86
Code Lines	27

Duplication

Lines	0
Ratio	0 %

Code Coverage

Tests	24
CRAP Score	15

Importance

Changes

0

# -*- coding: utf-8 -*-
def cmp_features(feat1, feat2, weights=None):
    """Compare features.

    This returns a number in the range [0, 1] representing a comparison of two
    feature bundles.

    If one of the bundles is negative, -1 is returned (for unknown values)

    If the bundles are identical, 1 is returned.

    If they are inverses of one another, 0 is returned.

    Otherwise, a float representing their similarity is returned.

    Parameters
    ----------
    feat1 : int
        A feature bundle
    feat2 : int
        A feature bundle
    weights : None or list or tuple or dict
        If None, all features are of equal significance and a simple normalized
        hamming distance of the features is calculated. If a list or tuple
        of numeric values is supplied, the values are inferred as the weights
        for each feature, in order of the features listed in _FEATURE_MASK.
        If a dict is supplied, its key values should match keys in
        _FEATURE_MASK to which each weight (value) should be assigned. Missing
        values in all cases are assigned a weight of 0 and will be omitted from
        the comparison.

    Returns
    -------
    float
        A comparison of the feature bundles

    Examples
    --------
    >>> cmp_features(ipa_to_features('l')[0], ipa_to_features('l')[0])
    1.0
    >>> cmp_features(ipa_to_features('l')[0], ipa_to_features('n')[0])
    0.8709677419354839
    >>> cmp_features(ipa_to_features('l')[0], ipa_to_features('z')[0])
    0.8709677419354839
    >>> cmp_features(ipa_to_features('l')[0], ipa_to_features('i')[0])
    0.564516129032258

    .. versionadded:: 0.1.0
    .. versionchanged:: 0.4.1
        Added weights parameter for modifiable feature weighting

    """
    if feat1 < 0 or feat2 < 0:
        return 0.0
    if feat1 == feat2:
        return 1.0

    # This should be handled some other way since this will take a long time
    # when done repeatedly. Maybe convert to a class & save the weights list.
    if weights is not None:
        if isinstance(weights, dict):
            weights = [
                weights[feature] if feature in weights else 0
                for feature in sorted(
                    _FEATURE_MASK, key=_FEATURE_MASK.get, reverse=True
                )
            ]
        elif isinstance(weights, (list, tuple)):
            weights = list(weights) + [0] * (len(_FEATURE_MASK) - len(weights))
        else:
            raise TypeError('weights must be a dist, list, or tuple.')

    magnitude = sum(weights) if weights else len(_FEATURE_MASK)

    featxor = feat1 ^ feat2
    diffbits = 0
    i = 0
    while featxor:
        if featxor & 0b1:
            diffbits += weights[i] if weights else 1
        featxor >>= 1
        if featxor & 0b1:
            diffbits += weights[i] if weights else 1
        featxor >>= 1
        i += 1
    return 1 - (0 if not diffbits else (diffbits / (2 * magnitude)))
    """
    diff_feats = 0