Skip to content

Оценки как внутренние компоненты детектора

Детекторы - это просто менеджеры над объектами оценок. Они представляют собой способы оценки каждой строки в каждой подвыборке.

Warning

Оценки имеют ограниченную поддержку типов данных.

applybn реализует две основные группы оценок: на основе модели и на основе близости.

Подробнее читайте в Руководстве пользователя.

Оценка

Bases: ABC

An abstract base class for implementing scoring mechanisms.

Source code in applybn/anomaly_detection/scores/score.py 
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
class Score(ABC):
    """
    An abstract base class for implementing scoring mechanisms.
    """

    def __init__(self, verbose: int = 1):
        """
        Initializes the Score object.

        Args:
            verbose: The verbosity level for logging. Default is 1.
        """
        self.verbose = verbose

    @abstractmethod
    def score(self, X: pd.DataFrame):
        """
        Abstract method to compute scores for the given input data.

        Args:
            X: The input data to be scored.

        Raises:
            NotImplementedError: This method must be implemented by subclasses.
        """
        pass

__init__(verbose=1)

Initializes the Score object.

Parameters:

Name Type Description Default
verbose int

The verbosity level for logging. Default is 1.

 1
Source code in applybn/anomaly_detection/scores/score.py 
10
11
12
13
14
15
16
17
def __init__(self, verbose: int = 1):
    """
    Initializes the Score object.

    Args:
        verbose: The verbosity level for logging. Default is 1.
    """
    self.verbose = verbose

score(X) abstractmethod

Abstract method to compute scores for the given input data.

Parameters:

Name Type Description Default
X DataFrame

The input data to be scored.

 required

Raises:

Type Description
NotImplementedError

This method must be implemented by subclasses.
Source code in applybn/anomaly_detection/scores/score.py 
19
20
21
22
23
24
25
26
27
28
29
30
@abstractmethod
def score(self, X: pd.DataFrame):
    """
    Abstract method to compute scores for the given input data.

    Args:
        X: The input data to be scored.

    Raises:
        NotImplementedError: This method must be implemented by subclasses.
    """
    pass

Оценки на основе близости

Оценка локальных выбросов

Bases: ProximityBasedScore

A class for computing outlier scores using the Local Outlier Factor (LOF) algorithm.

Source code in applybn/anomaly_detection/scores/proximity_based.py 
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
class LocalOutlierScore(ProximityBasedScore):
    """
    A class for computing outlier scores using the Local Outlier Factor (LOF) algorithm.
    """

    def __init__(self, proximity_steps: int = 5, verbose: int = 1, **kwargs):
        """
        Initializes the LocalOutlierScore object.

        Args:
            proximity_steps: The number of proximity steps to perform. Default is 5.
            verbose: The verbosity level for logging. Default is 1.
            **kwargs: Additional parameters for the Local Outlier Factor algorithm.
        """
        super().__init__(proximity_steps=proximity_steps, verbose=verbose)
        self.params = kwargs

    def local_score(self, X: pd.DataFrame):
        """
        Computes the local outlier scores for the given data using the LOF algorithm.

        Args:
            X: The input data.

        Returns:
            np.ndarray: An array of negative outlier factors, where higher values indicate more abnormal data points.
        """
        clf = LocalOutlierFactor(**self.params)
        clf.fit(X)
        # The higher the value, the more abnormal the data point
        return np.negative(clf.negative_outlier_factor_)

__init__(proximity_steps=5, verbose=1, **kwargs)

Initializes the LocalOutlierScore object.

Parameters:

Name Type Description Default
proximity_steps int

The number of proximity steps to perform. Default is 5.

 5
verbose int

The verbosity level for logging. Default is 1.

 1
**kwargs

Additional parameters for the Local Outlier Factor algorithm.

 {}
Source code in applybn/anomaly_detection/scores/proximity_based.py 
72
73
74
75
76
77
78
79
80
81
82
def __init__(self, proximity_steps: int = 5, verbose: int = 1, **kwargs):
    """
    Initializes the LocalOutlierScore object.

    Args:
        proximity_steps: The number of proximity steps to perform. Default is 5.
        verbose: The verbosity level for logging. Default is 1.
        **kwargs: Additional parameters for the Local Outlier Factor algorithm.
    """
    super().__init__(proximity_steps=proximity_steps, verbose=verbose)
    self.params = kwargs

local_score(X)

Computes the local outlier scores for the given data using the LOF algorithm.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required

Returns:

Type Description

np.ndarray: An array of negative outlier factors, where higher values indicate more abnormal data points.
Source code in applybn/anomaly_detection/scores/proximity_based.py 
84
85
86
87
88
89
90
91
92
93
94
95
96
97
def local_score(self, X: pd.DataFrame):
    """
    Computes the local outlier scores for the given data using the LOF algorithm.

    Args:
        X: The input data.

    Returns:
        np.ndarray: An array of negative outlier factors, where higher values indicate more abnormal data points.
    """
    clf = LocalOutlierFactor(**self.params)
    clf.fit(X)
    # The higher the value, the more abnormal the data point
    return np.negative(clf.negative_outlier_factor_)

Оценка на основе Isolation Forest

Bases: ProximityBasedScore

A class for computing outlier scores using the Isolation Forest algorithm.

Source code in applybn/anomaly_detection/scores/proximity_based.py 
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
class IsolationForestScore(ProximityBasedScore):
    """
    A class for computing outlier scores using the Isolation Forest algorithm.
    """

    def __init__(self, proximity_steps: int = 5, verbose: int = 1, **kwargs):
        """
        Initializes the IsolationForestScore object.

        Args:
            **kwargs: Additional parameters for the Isolation Forest algorithm.
        """
        super().__init__(verbose=verbose, proximity_steps=proximity_steps)
        self.params = kwargs

    def local_score(self, X: pd.DataFrame):
        """
        Computes the outlier scores for the given data using the Isolation Forest algorithm.

        Args:
            X: The input data.

        Returns:
            np.ndarray: An array of negative decision function values, where higher values indicate more abnormal data points.
        """
        clf = IsolationForest(**self.params)
        clf.fit(X)
        return np.negative(clf.decision_function(X))

__init__(proximity_steps=5, verbose=1, **kwargs)

Initializes the IsolationForestScore object.

Parameters:

Name Type Description Default
**kwargs

Additional parameters for the Isolation Forest algorithm.

 {}
Source code in applybn/anomaly_detection/scores/proximity_based.py 
105
106
107
108
109
110
111
112
113
def __init__(self, proximity_steps: int = 5, verbose: int = 1, **kwargs):
    """
    Initializes the IsolationForestScore object.

    Args:
        **kwargs: Additional parameters for the Isolation Forest algorithm.
    """
    super().__init__(verbose=verbose, proximity_steps=proximity_steps)
    self.params = kwargs

local_score(X)

Computes the outlier scores for the given data using the Isolation Forest algorithm.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required

Returns:

Type Description

np.ndarray: An array of negative decision function values, where higher values indicate more abnormal data points.
Source code in applybn/anomaly_detection/scores/proximity_based.py 
115
116
117
118
119
120
121
122
123
124
125
126
127
def local_score(self, X: pd.DataFrame):
    """
    Computes the outlier scores for the given data using the Isolation Forest algorithm.

    Args:
        X: The input data.

    Returns:
        np.ndarray: An array of negative decision function values, where higher values indicate more abnormal data points.
    """
    clf = IsolationForest(**self.params)
    clf.fit(X)
    return np.negative(clf.decision_function(X))

Оценки на основе модели

Bases: Score

A generic score class that computes scores based on a provided model. Model must implement the predict_proba method.

Source code in applybn/anomaly_detection/scores/model_based.py 
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
class ModelBasedScore(Score):
    """
    A generic score class that computes scores based on a provided model.
    Model must implement the predict_proba method.
    """

    def __init__(self, model):
        """
        Initializes the ModelBasedScore object.

        Args:
            model: The model used to compute probabilities for scoring.
        """
        super().__init__()
        self.model = model

    def score(self, X) -> np.ndarray:
        """
        Computes the score for the input data using the model's predicted probabilities.

        Args:
            X: The input data to be scored.

        Returns:
            np.ndarray: The predicted probabilities for the input data.
        """
        if not hasattr(self.model, "predict_proba"):
            raise AttributeError("The model does not have a predict_proba method.")
        probas = self.model.predict_proba(X)

        if isinstance(probas, pd.Series):
            return probas.values
        if isinstance(probas, np.ndarray):
            return probas

__init__(model)

Initializes the ModelBasedScore object.

Parameters:

Name Type Description Default
model

The model used to compute probabilities for scoring.

 required
Source code in applybn/anomaly_detection/scores/model_based.py 
15
16
17
18
19
20
21
22
23
def __init__(self, model):
    """
    Initializes the ModelBasedScore object.

    Args:
        model: The model used to compute probabilities for scoring.
    """
    super().__init__()
    self.model = model

score(X)

Computes the score for the input data using the model's predicted probabilities.

Parameters:

Name Type Description Default
X

The input data to be scored.

 required

Returns:

Type Description
ndarray

np.ndarray: The predicted probabilities for the input data.
Source code in applybn/anomaly_detection/scores/model_based.py 
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
def score(self, X) -> np.ndarray:
    """
    Computes the score for the input data using the model's predicted probabilities.

    Args:
        X: The input data to be scored.

    Returns:
        np.ndarray: The predicted probabilities for the input data.
    """
    if not hasattr(self.model, "predict_proba"):
        raise AttributeError("The model does not have a predict_proba method.")
    probas = self.model.predict_proba(X)

    if isinstance(probas, pd.Series):
        return probas.values
    if isinstance(probas, np.ndarray):
        return probas

Оценка на основе BN

Bases: Score

A score class based on a Bayesian network (BN).

Attributes:

Name Type Description
bn

The Bayesian network used for scoring.
encoding

The encoding for discrete variables.
child_nodes

The child nodes in the Bayesian network.
verbose

The verbosity level for logging.
Source code in applybn/anomaly_detection/scores/model_based.py 
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
class BNBasedScore(Score):
    """
    A score class based on a Bayesian network (BN).

    Attributes:
        bn: The Bayesian network used for scoring.
        encoding: The encoding for discrete variables.
        child_nodes: The child nodes in the Bayesian network.
        verbose: The verbosity level for logging.
    """

    def __init__(self, bn: bamt_network, encoding: dict, verbose: int = 1):
        """
        Initializes the BNBasedScore object.

        Args:
            bn: The Bayesian network used for scoring.
            encoding: The encoding for discrete variables.
            verbose: The verbosity level for logging.
        """
        super().__init__(verbose=verbose)
        self.encoding = encoding
        self.bn = bn

        child_nodes = []
        for column in bn.nodes_names:
            if self.bn[column].disc_parents + self.bn[column].cont_parents:
                child_nodes.append(column)

        self.child_nodes = child_nodes

    def local_score(self, X: pd.DataFrame, node_name: str):
        """
        Computes the local score for a specific node in the Bayesian network.

        Args:
            X: The input data.
            node_name : The name of the node to compute the score for.

        Returns:
            np.ndarray: An array of local scores for the specified node.
        """
        node = self.bn[node_name]
        diff = []
        parents = node.cont_parents + node.disc_parents
        parent_dtypes = X[parents].dtypes.to_dict()

        for i in X.index:
            node_value = X.loc[i, node_name]
            row_df = X.loc[[i], parents].astype(parent_dtypes)
            pvalues = row_df.to_dict("records")[0]
            cond_dist = self.bn.get_dist(node_name, pvals=pvalues)

            if "gaussian" in cond_dist.node_type:
                cond_mean, std = cond_dist.get()
            else:
                probs, classes = cond_dist.get()
                match self.bn.descriptor["types"][node_name]:
                    case "disc_num":
                        classes_ = [int(class_name) for class_name in classes]
                    case "disc":
                        classes_ = np.asarray(
                            [
                                self.encoding[node_name][class_name]
                                for class_name in classes
                            ]
                        )
                cond_mean = classes_ @ np.asarray(probs).T

            match self.bn.descriptor["types"][node_name]:
                case "disc_num":
                    diff.append((node_value - cond_mean))
                case "disc":
                    diff.append(self.encoding[node_name][node_value] - cond_mean)
                case "cont":
                    diff.append((node_value - cond_mean) / std)

        return np.asarray(diff).reshape(-1, 1)

    def score(self, X: pd.DataFrame):
        """
        Computes the scores for all child nodes in the Bayesian network.

        Args:
            X: The input data.

        Returns:
            np.ndarray: A 2D array of scores for all child nodes.
        """
        if self.verbose >= 1:
            model_iterator = tqdm(self.child_nodes, desc="Model")
        else:
            model_iterator = self.child_nodes

        model_factors = []
        for child_node in model_iterator:
            model_factors.append(self.local_score(X, child_node))

        return np.hstack(model_factors)

__init__(bn, encoding, verbose=1)

Initializes the BNBasedScore object.

Parameters:

Name Type Description Default
bn bamt_network

The Bayesian network used for scoring.

 required
encoding dict

The encoding for discrete variables.

 required
verbose int

The verbosity level for logging.

 1
Source code in applybn/anomaly_detection/scores/model_based.py 
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
def __init__(self, bn: bamt_network, encoding: dict, verbose: int = 1):
    """
    Initializes the BNBasedScore object.

    Args:
        bn: The Bayesian network used for scoring.
        encoding: The encoding for discrete variables.
        verbose: The verbosity level for logging.
    """
    super().__init__(verbose=verbose)
    self.encoding = encoding
    self.bn = bn

    child_nodes = []
    for column in bn.nodes_names:
        if self.bn[column].disc_parents + self.bn[column].cont_parents:
            child_nodes.append(column)

    self.child_nodes = child_nodes

local_score(X, node_name)

Computes the local score for a specific node in the Bayesian network.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required
node_name

The name of the node to compute the score for.

 required

Returns:

Type Description

np.ndarray: An array of local scores for the specified node.
Source code in applybn/anomaly_detection/scores/model_based.py 
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
def local_score(self, X: pd.DataFrame, node_name: str):
    """
    Computes the local score for a specific node in the Bayesian network.

    Args:
        X: The input data.
        node_name : The name of the node to compute the score for.

    Returns:
        np.ndarray: An array of local scores for the specified node.
    """
    node = self.bn[node_name]
    diff = []
    parents = node.cont_parents + node.disc_parents
    parent_dtypes = X[parents].dtypes.to_dict()

    for i in X.index:
        node_value = X.loc[i, node_name]
        row_df = X.loc[[i], parents].astype(parent_dtypes)
        pvalues = row_df.to_dict("records")[0]
        cond_dist = self.bn.get_dist(node_name, pvals=pvalues)

        if "gaussian" in cond_dist.node_type:
            cond_mean, std = cond_dist.get()
        else:
            probs, classes = cond_dist.get()
            match self.bn.descriptor["types"][node_name]:
                case "disc_num":
                    classes_ = [int(class_name) for class_name in classes]
                case "disc":
                    classes_ = np.asarray(
                        [
                            self.encoding[node_name][class_name]
                            for class_name in classes
                        ]
                    )
            cond_mean = classes_ @ np.asarray(probs).T

        match self.bn.descriptor["types"][node_name]:
            case "disc_num":
                diff.append((node_value - cond_mean))
            case "disc":
                diff.append(self.encoding[node_name][node_value] - cond_mean)
            case "cont":
                diff.append((node_value - cond_mean) / std)

    return np.asarray(diff).reshape(-1, 1)

score(X)

Computes the scores for all child nodes in the Bayesian network.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required

Returns:

Type Description

np.ndarray: A 2D array of scores for all child nodes.
Source code in applybn/anomaly_detection/scores/model_based.py 
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
def score(self, X: pd.DataFrame):
    """
    Computes the scores for all child nodes in the Bayesian network.

    Args:
        X: The input data.

    Returns:
        np.ndarray: A 2D array of scores for all child nodes.
    """
    if self.verbose >= 1:
        model_iterator = tqdm(self.child_nodes, desc="Model")
    else:
        model_iterator = self.child_nodes

    model_factors = []
    for child_node in model_iterator:
        model_factors.append(self.local_score(X, child_node))

    return np.hstack(model_factors)

Оценка на основе IQR

Bases: BNBasedScore

A score class that uses the Interquartile Range (IQR) for anomaly detection.

Source code in applybn/anomaly_detection/scores/model_based.py 
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
class IQRBasedScore(BNBasedScore):
    """
    A score class that uses the Interquartile Range (IQR) for anomaly detection.
    """

    def __init__(
        self,
        bn: bamt_network,
        encoding: dict,
        iqr_sensivity: float = 1.0,
        verbose: int = 1,
    ):
        """
        Initializes the IQRBasedScore object.

        Args:
            bn: The Bayesian network used for scoring.
            encoding: The encoding for discrete variables.
            iqr_sensivity: The sensitivity factor for IQR-based scoring.
            verbose: The verbosity level for logging.
        """
        super().__init__(bn=bn, encoding=encoding, verbose=verbose)
        self.iqr_sensivity = iqr_sensivity

    @staticmethod
    def score_iqr(
        upper: float, lower: float, y: float, max_distance: float, min_distance: float
    ):
        """
        Computes the IQR-based score for a given value.

        Args:
            upper: The upper bound of the IQR.
            lower: The lower bound of the IQR.
            y: The value to score.
            max_distance: The maximum distance for scaling.
            min_distance: The minimum distance for scaling.

        Raises:
            ValueError: If the closest value does not match either upper or lower bound.

        Returns:
            float: The IQR-based score.
        """
        if lower < y <= upper:
            return 0

        closest_value = min([upper, lower], key=lambda x: abs(x - y))
        current_distance = abs(closest_value - y)

        if closest_value == upper:
            ref_distance = max_distance
        elif closest_value == lower:
            ref_distance = min_distance
        else:
            raise ValueError(
                "Unexpected state: closest_value does not match either upper or lower bound."
            )

        return min(1, current_distance / abs(ref_distance))

    def local_score(self, X: pd.DataFrame, node_name: str):
        """
        Computes the local IQR-based score for a specific node.

        Args:
            X: The input data.
            node_name : The name of the node to compute the score for.

        Returns:
            np.ndarray: An array of local scores for the specified node.
        """
        node = self.bn[node_name]
        parents = node.cont_parents + node.disc_parents
        parent_dtypes = X[parents].dtypes.to_dict()

        scores = []
        for i in X.index:
            row_df = X.loc[[i], parents].astype(parent_dtypes)
            pvalues = row_df.to_dict("records")[0]
            dist = self.bn.get_dist(node_name, pvals=pvalues).get(with_gaussian=True)

            X_value = X.loc[i, node_name]
            q25 = dist.ppf(0.25)
            q75 = dist.ppf(0.75)
            iqr = q75 - q25

            lower_bound = q25 - iqr * self.iqr_sensivity
            upper_bound = q75 + iqr * self.iqr_sensivity

            scores.append(
                self.score_iqr(
                    upper_bound,
                    lower_bound,
                    X_value,
                    max_distance=1 * X[node_name].max(),
                    min_distance=1 * X[node_name].min(),
                )
            )

        return np.asarray(scores).reshape(-1, 1)

__init__(bn, encoding, iqr_sensivity=1.0, verbose=1)

Initializes the IQRBasedScore object.

Parameters:

Name Type Description Default
bn bamt_network

The Bayesian network used for scoring.

 required
encoding dict

The encoding for discrete variables.

 required
iqr_sensivity float

The sensitivity factor for IQR-based scoring.

 1.0
verbose int

The verbosity level for logging.

 1
Source code in applybn/anomaly_detection/scores/model_based.py 
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
def __init__(
    self,
    bn: bamt_network,
    encoding: dict,
    iqr_sensivity: float = 1.0,
    verbose: int = 1,
):
    """
    Initializes the IQRBasedScore object.

    Args:
        bn: The Bayesian network used for scoring.
        encoding: The encoding for discrete variables.
        iqr_sensivity: The sensitivity factor for IQR-based scoring.
        verbose: The verbosity level for logging.
    """
    super().__init__(bn=bn, encoding=encoding, verbose=verbose)
    self.iqr_sensivity = iqr_sensivity

local_score(X, node_name)

Computes the local IQR-based score for a specific node.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required
node_name

The name of the node to compute the score for.

 required

Returns:

Type Description

np.ndarray: An array of local scores for the specified node.
Source code in applybn/anomaly_detection/scores/model_based.py 
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
def local_score(self, X: pd.DataFrame, node_name: str):
    """
    Computes the local IQR-based score for a specific node.

    Args:
        X: The input data.
        node_name : The name of the node to compute the score for.

    Returns:
        np.ndarray: An array of local scores for the specified node.
    """
    node = self.bn[node_name]
    parents = node.cont_parents + node.disc_parents
    parent_dtypes = X[parents].dtypes.to_dict()

    scores = []
    for i in X.index:
        row_df = X.loc[[i], parents].astype(parent_dtypes)
        pvalues = row_df.to_dict("records")[0]
        dist = self.bn.get_dist(node_name, pvals=pvalues).get(with_gaussian=True)

        X_value = X.loc[i, node_name]
        q25 = dist.ppf(0.25)
        q75 = dist.ppf(0.75)
        iqr = q75 - q25

        lower_bound = q25 - iqr * self.iqr_sensivity
        upper_bound = q75 + iqr * self.iqr_sensivity

        scores.append(
            self.score_iqr(
                upper_bound,
                lower_bound,
                X_value,
                max_distance=1 * X[node_name].max(),
                min_distance=1 * X[node_name].min(),
            )
        )

    return np.asarray(scores).reshape(-1, 1)

score_iqr(upper, lower, y, max_distance, min_distance) staticmethod

Computes the IQR-based score for a given value.

Parameters:

Name Type Description Default
upper float

The upper bound of the IQR.

 required
lower float

The lower bound of the IQR.

 required
y float

The value to score.

 required
max_distance float

The maximum distance for scaling.

 required
min_distance float

The minimum distance for scaling.

 required

Raises:

Type Description
ValueError

If the closest value does not match either upper or lower bound.

Returns:

Name Type Description
float

The IQR-based score.
Source code in applybn/anomaly_detection/scores/model_based.py 
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
@staticmethod
def score_iqr(
    upper: float, lower: float, y: float, max_distance: float, min_distance: float
):
    """
    Computes the IQR-based score for a given value.

    Args:
        upper: The upper bound of the IQR.
        lower: The lower bound of the IQR.
        y: The value to score.
        max_distance: The maximum distance for scaling.
        min_distance: The minimum distance for scaling.

    Raises:
        ValueError: If the closest value does not match either upper or lower bound.

    Returns:
        float: The IQR-based score.
    """
    if lower < y <= upper:
        return 0

    closest_value = min([upper, lower], key=lambda x: abs(x - y))
    current_distance = abs(closest_value - y)

    if closest_value == upper:
        ref_distance = max_distance
    elif closest_value == lower:
        ref_distance = min_distance
    else:
        raise ValueError(
            "Unexpected state: closest_value does not match either upper or lower bound."
        )

    return min(1, current_distance / abs(ref_distance))

Оценка на основе отношения условных вероятностей

Bases: BNBasedScore

A score class that uses conditional probability ratios for anomaly detection.

Source code in applybn/anomaly_detection/scores/model_based.py 
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
class CondRatioScore(BNBasedScore):
    """
    A score class that uses conditional probability ratios for anomaly detection.
    """

    def __init__(self, bn: bamt_network, encoding: dict, verbose: int = 1):
        """
        Initializes the CondRatioScore object.

        Args:
            bn: The Bayesian network used for scoring.
            encoding: The encoding for discrete variables.
            verbose: The verbosity level for logging.
        """
        super(CondRatioScore, self).__init__(bn=bn, encoding=encoding, verbose=verbose)

    def local_score(self, X: pd.DataFrame, node_name: str):
        """
        Computes the local conditional ratio score for a specific node.

        Args:
            X: The input data.
            node_name: The name of the node to compute the score for.

        Returns:
            np.ndarray: An array of local scores for the specified node.
        """
        node = self.bn[node_name]
        diff = []
        parents = node.cont_parents + node.disc_parents
        parent_dtypes = X[parents].dtypes.to_dict()

        for i in X.index:
            row_df = X.loc[[i], parents].astype(parent_dtypes)
            pvalues = row_df.to_dict("records")[0]
            node_value = X.loc[i, node_name]
            cond_dist = self.bn.get_dist(node_name, pvals=pvalues).get()

            diff.append(self.score_proba_ratio(X[node_name], node_value, cond_dist))

        return np.asarray(diff).reshape(-1, 1)

    @staticmethod
    def score_proba_ratio(sample: pd.Series, X_value: str, cond_dist: tuple):
        """
        Computes the conditional probability ratio score.

        Args:
            sample: The sample data.
            X_value: The value to score.
            cond_dist: The conditional distribution.

        Returns:
            float: The conditional probability ratio score.
        """
        cond_probs, values = cond_dist
        marginal_prob = sample.value_counts(normalize=True)[X_value]

        index = values.index(str(X_value))
        cond_prob = cond_probs[index]

        if not np.isfinite(marginal_prob / cond_prob):
            return np.nan

        return min(1, marginal_prob / cond_prob)

__init__(bn, encoding, verbose=1)

Initializes the CondRatioScore object.

Parameters:

Name Type Description Default
bn bamt_network

The Bayesian network used for scoring.

 required
encoding dict

The encoding for discrete variables.

 required
verbose int

The verbosity level for logging.

 1
Source code in applybn/anomaly_detection/scores/model_based.py 
254
255
256
257
258
259
260
261
262
263
def __init__(self, bn: bamt_network, encoding: dict, verbose: int = 1):
    """
    Initializes the CondRatioScore object.

    Args:
        bn: The Bayesian network used for scoring.
        encoding: The encoding for discrete variables.
        verbose: The verbosity level for logging.
    """
    super(CondRatioScore, self).__init__(bn=bn, encoding=encoding, verbose=verbose)

local_score(X, node_name)

Computes the local conditional ratio score for a specific node.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required
node_name str

The name of the node to compute the score for.

 required

Returns:

Type Description

np.ndarray: An array of local scores for the specified node.
Source code in applybn/anomaly_detection/scores/model_based.py 
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
def local_score(self, X: pd.DataFrame, node_name: str):
    """
    Computes the local conditional ratio score for a specific node.

    Args:
        X: The input data.
        node_name: The name of the node to compute the score for.

    Returns:
        np.ndarray: An array of local scores for the specified node.
    """
    node = self.bn[node_name]
    diff = []
    parents = node.cont_parents + node.disc_parents
    parent_dtypes = X[parents].dtypes.to_dict()

    for i in X.index:
        row_df = X.loc[[i], parents].astype(parent_dtypes)
        pvalues = row_df.to_dict("records")[0]
        node_value = X.loc[i, node_name]
        cond_dist = self.bn.get_dist(node_name, pvals=pvalues).get()

        diff.append(self.score_proba_ratio(X[node_name], node_value, cond_dist))

    return np.asarray(diff).reshape(-1, 1)

score_proba_ratio(sample, X_value, cond_dist) staticmethod

Computes the conditional probability ratio score.

Parameters:

Name Type Description Default
sample Series

The sample data.

 required
X_value str

The value to score.

 required
cond_dist tuple

The conditional distribution.

 required

Returns:

Name Type Description
float

The conditional probability ratio score.
Source code in applybn/anomaly_detection/scores/model_based.py 
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
@staticmethod
def score_proba_ratio(sample: pd.Series, X_value: str, cond_dist: tuple):
    """
    Computes the conditional probability ratio score.

    Args:
        sample: The sample data.
        X_value: The value to score.
        cond_dist: The conditional distribution.

    Returns:
        float: The conditional probability ratio score.
    """
    cond_probs, values = cond_dist
    marginal_prob = sample.value_counts(normalize=True)[X_value]

    index = values.index(str(X_value))
    cond_prob = cond_probs[index]

    if not np.isfinite(marginal_prob / cond_prob):
        return np.nan

    return min(1, marginal_prob / cond_prob)

Комбинированная оценка на основе IQR и отношения условных вероятностей

Bases: BNBasedScore

A score class that combines IQR-based scoring and probability ratio scoring for anomaly detection.

Source code in applybn/anomaly_detection/scores/model_based.py 
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
class CombinedIQRandProbRatioScore(BNBasedScore):
    """
    A score class that combines IQR-based scoring and probability ratio scoring for anomaly detection.
    """

    def __init__(
        self, bn: bamt_network, encoding: dict, scores: dict, verbose: int = 1
    ):
        """
        Initializes the CombinedIQRandProbRatioScore object.

        Args:
            bn: The Bayesian network used for scoring.
            encoding: The encoding for discrete variables.
            scores: A dictionary containing scoring objects for continuous and discrete variables.
            verbose: The verbosity level for logging.
        """
        super(CombinedIQRandProbRatioScore, self).__init__(
            bn=bn, encoding=encoding, verbose=verbose
        )
        self.scores = scores

    def local_score(self, X: pd.DataFrame, node_name: str):
        """
        Computes the local score for a specific node by combining IQR-based and probability ratio scoring.

        Args:
            X: The input data.
            node_name: The name of the node to compute the score for.

        Returns:
            np.ndarray: An array of local scores for the specified node.
        """
        node = self.bn[node_name]
        iqr_sensivity = self.scores["cont"].iqr_sensivity
        parents = node.cont_parents + node.disc_parents
        parent_dtypes = X[parents].dtypes.to_dict()

        scores = []
        for i in X.index:
            row_df = X.loc[[i], parents].astype(parent_dtypes)
            pvalues = row_df.to_dict("records")[0]
            X_value = X.loc[i, node_name]
            dist = self.bn.get_dist(node_name, pvals=pvalues)

            if "gaussian" in dist.node_type:
                dist = dist.get(with_gaussian=True)
                if dist.kwds["scale"] == 0:
                    scores.append(0)
                    continue

                q25 = dist.ppf(0.25)
                q75 = dist.ppf(0.75)
                iqr = q75 - q25

                lower_bound = q25 - iqr * iqr_sensivity
                upper_bound = q75 + iqr * iqr_sensivity

                scores.append(
                    self.scores["cont"].score_iqr(
                        upper_bound,
                        lower_bound,
                        X_value,
                        max_distance=1 * X[node_name].max(),
                        min_distance=1 * X[node_name].min(),
                    )
                )
            else:
                dist = dist.get()
                scores.append(
                    self.scores["disc"].score_proba_ratio(X[node_name], X_value, dist)
                )

        return np.asarray(scores).reshape(-1, 1)

__init__(bn, encoding, scores, verbose=1)

Initializes the CombinedIQRandProbRatioScore object.

Parameters:

Name Type Description Default
bn bamt_network

The Bayesian network used for scoring.

 required
encoding dict

The encoding for discrete variables.

 required
scores dict

A dictionary containing scoring objects for continuous and discrete variables.

 required
verbose int

The verbosity level for logging.

 1
Source code in applybn/anomaly_detection/scores/model_based.py 
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
def __init__(
    self, bn: bamt_network, encoding: dict, scores: dict, verbose: int = 1
):
    """
    Initializes the CombinedIQRandProbRatioScore object.

    Args:
        bn: The Bayesian network used for scoring.
        encoding: The encoding for discrete variables.
        scores: A dictionary containing scoring objects for continuous and discrete variables.
        verbose: The verbosity level for logging.
    """
    super(CombinedIQRandProbRatioScore, self).__init__(
        bn=bn, encoding=encoding, verbose=verbose
    )
    self.scores = scores

local_score(X, node_name)

Computes the local score for a specific node by combining IQR-based and probability ratio scoring.

Parameters:

Name Type Description Default
X DataFrame

The input data.

 required
node_name str

The name of the node to compute the score for.

 required

Returns:

Type Description

np.ndarray: An array of local scores for the specified node.
Source code in applybn/anomaly_detection/scores/model_based.py 
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
def local_score(self, X: pd.DataFrame, node_name: str):
    """
    Computes the local score for a specific node by combining IQR-based and probability ratio scoring.

    Args:
        X: The input data.
        node_name: The name of the node to compute the score for.

    Returns:
        np.ndarray: An array of local scores for the specified node.
    """
    node = self.bn[node_name]
    iqr_sensivity = self.scores["cont"].iqr_sensivity
    parents = node.cont_parents + node.disc_parents
    parent_dtypes = X[parents].dtypes.to_dict()

    scores = []
    for i in X.index:
        row_df = X.loc[[i], parents].astype(parent_dtypes)
        pvalues = row_df.to_dict("records")[0]
        X_value = X.loc[i, node_name]
        dist = self.bn.get_dist(node_name, pvals=pvalues)

        if "gaussian" in dist.node_type:
            dist = dist.get(with_gaussian=True)
            if dist.kwds["scale"] == 0:
                scores.append(0)
                continue

            q25 = dist.ppf(0.25)
            q75 = dist.ppf(0.75)
            iqr = q75 - q25

            lower_bound = q25 - iqr * iqr_sensivity
            upper_bound = q75 + iqr * iqr_sensivity

            scores.append(
                self.scores["cont"].score_iqr(
                    upper_bound,
                    lower_bound,
                    X_value,
                    max_distance=1 * X[node_name].max(),
                    min_distance=1 * X[node_name].min(),
                )
            )
        else:
            dist = dist.get()
            scores.append(
                self.scores["disc"].score_proba_ratio(X[node_name], X_value, dist)
            )

    return np.asarray(scores).reshape(-1, 1)