lore.py

import json
from numpy import finfo, inf
from models import Rule


def lore_to_glocalx(json_file, info_file, class_name='class'):
    """Load file `json_file` and `info_file` and return the loaded JSON
    rules in a list of `Rule` objects preserving the loading order.
    Args:
        json_file (str): Path to the JSON file.
        info_file (str): Path to the info file containing the rules' metadata.
        class_name (str): Name of the class feature found in the rules. Defaults to 'class'
    Returns:
        (list): List of `Rule` objects.
    """
    with open(json_file, 'r') as rules_log, open(info_file, 'r') as info_log:
        rules = json.load(rules_log)
        infos = json.load(info_log)
    class_values = infos['class_values']
    feature_names = infos['feature_names']

    rules = [r for r in rules if len(r) > 0]
    output_rules = []
    for rule in rules:
        # rules are [consequence, premises, infos]
        if len(rule) != 3 or class_name not in rule[0]:
            continue
        try:
            consequence = class_values.index(rule[0][class_name])
        except KeyError as e:
            raise e
        actual_premises = dict()
        for f, val in rule[1].items():
            # lower and upper bound, add the check on feature for categorical values with '<' and '<='
            if '< ' in val and ' <=' in val and f in feature_names:
                f_idx = feature_names.index(f)
                actual_premises[f_idx] = (float(val.split('< ')[0]), float(val.split(' <=')[1]))
            # lower but no upper bound
            elif val.startswith('>') and val[1] != '=':
                f_idx = feature_names.index(f)
                actual_premises[f_idx] = (float(val.split('>')[1]), inf)
            # no lower but upper bound
            elif val.startswith('<='):
                f_idx = feature_names.index(f)
                actual_premises[f_idx] = (-inf, float(val.split('<=')[1]))
            # one-hot
            elif ('<' not in val and '<=' not in val) or ('< ' in val and ' <=' in val and f not in feature_names):
                if f + '=' + val in feature_names:
                    f_idx = feature_names.index(f + '=' + val)
                    actual_premises[f_idx] = (0.5, 1 + finfo(float).eps)
                else:
                    f_idx = feature_names.index(f)
                    actual_premises[f_idx] = (float(val) - finfo(float).eps, float(val))

        transformed_rule = Rule(premises=actual_premises, consequence=consequence, names=feature_names)
        output_rules.append(transformed_rule)

    output_rules = set(output_rules)

    return output_rules