def flag_anomaly(self, X) -> np.ndarray:
"""
Flag the data point as an anomaly if the probability surpass epsilon (threshold).
Args:
X (numpy.ndarray or pandas.DataFrame): data to flag as an anomaly or not.
Returns:
numpy.ndarray: list of bool telling if a data point is an anomaly or not.
"""
if not self.check_if_trained():
raise ModelNotTrained("The model has not been trained!")
probabilities = self.predict(X)
return probabilities < self._epsilon
def flag_anomaly(self, X) -> np.ndarray:
"""
Flags as anomaly or not the data points.
Args:
X (numpy.ndarray or pandas.DataFrame): data.
Returns:
np.ndarray: list of bool telling if a data point is an anomaly or not.
"""
if not self.check_if_trained():
raise ModelNotTrained("The model has not been trained!")
scores = self.predict(X)
return scores > self._threshold
def flag_anomaly(self, X) -> np.ndarray:
"""
Flag a data point as an anomaly or as an inlier. If the score from the predict
method is negative, then it's an anomaly, if it's positive then it's an inlier.
Args:
X (numpy.ndarray or pandas.DataFrame): data to be flagged.
Returns:
numpy.ndarray: list containing bool values telling if data point is an
anomaly or not.
"""
if not self.check_if_trained():
raise ModelNotTrained("The model has not been trained!")
scores = self.predict(X)
return scores < 0
def flag_anomaly(self, X) -> np.ndarray:
"""
Flag the data points as anomaly if the calculated Mahalanobis distance surpass a
established threshold.
Args:
X (numpy.ndarray or pandas.DataFrame): data to flag as anomalous or not
anomalous.
Returns:
numpy.ndarray: list of booleans telling if point is anomalous or not.
"""
if not self.check_if_trained():
raise ModelNotTrained("The model has not been trained!")
distances = self.predict(X)
return distances > self._threshold
def flag_anomaly(self, X) -> np.ndarray:
"""
Flag a data point as an anomaly if the MAE (Mean Absolute Error) is higher than
the established threshold.
Args:
X (numpy.ndarray or pandas.DataFrame): data
Returns:
numpy.ndarray: list containing bool values telling if data point is an
anomaly or not.
"""
if not self.check_if_trained():
raise ModelNotTrained("The model has not been trained!")
predict = self._autoencoder.predict(X)
loss = self.mean_absolute_error(X, predict)
return loss > self._threshold