outlier.py

from sklearn.cluster import DBSCAN

class OutlierDetector(object):
    """Detect outlier users
    
    Parameters
    ----------
    users : [User]
    outlier_detector : DBScan or IsolationForest or other outlier detector or None
        Defaults to DBScan
        
    """
    
    def __init__(self, users, outlier_detector=None):
        self._users = users
        self._policies = list({pol for u in users for pol in u.policies})
        if outlier_detector is None:
            outlier_detector = DBSCAN(
                eps = 1,
                metric="euclidean",
                min_samples = 3,
                n_jobs = -1) 
        self._outlier_detector = outlier_detector
        self._user_policies = np.array([self._policy_vec(u) for u in self._users])
        
    def _policy_vec(self, user):
        return np.array([pol in user.policies for pol in self._policies], dtype=float) 
    
    def current_outliers(self):
        """Return outliers among input users
        
        Returns
        -------
        cluster_vals : [int]
            Negative values correspond to outlier indexes among users.
            
        """
        return self._outlier_detector.fit_predict(self._user_policies)

det = OutlierDetector(users)
outliers = det.current_outliers()
print(outliers)
print([user for i, user in enumerate(users) if outliers[i] < 0])