Utils for Anomaly Detection

Common functions used by anomaly_model.py, scorers.py, aggregators.py and detectors.py

darts.ad.utils.eval_accuracy_from_binary_prediction(actual_anomalies, binary_pred_anomalies, window=1, metric='recall')[source]

Score the results against true anomalies.

checks that pred_anomalies and actual_anomalies are the same:

  • type,

  • length,

  • number of components

  • binary and has values belonging to the two classes (1 and 0)

If one series is given for actual_anomalies and pred_anomalies contains more than one series, the function will consider actual_anomalies as the true anomalies for all scores in anomaly_score.

Parameters

  • actual_anomalies (Union[TimeSeries, Sequence[TimeSeries]]) – The (sequence of) ground truth of the anomalies (1 if it is an anomaly and 0 if not)

  • binary_pred_anomalies (Union[TimeSeries, Sequence[TimeSeries]]) – Anomaly predictions.

  • window (Union[int, Sequence[int]]) – Integer value indicating the number of past samples each point represents in the pred_anomalies. The parameter will be used to transform actual_anomalies. If a list is given. the length must match the number of series in pred_anomalies and actual_anomalies. If only one window is given, the value will be used for every series in pred_anomalies and actual_anomalies.

  • metric (str) – Optionally, Scoring function to use. Must be one of “recall”, “precision”, “f1”, and “accuracy”. Default: “recall”

Returns

Score of the anomalies prediction

  • float if binary_pred_anomalies is a univariate series (dimension=1).

  • Sequence[float]

    • if binary_pred_anomalies is a multivariate series (dimension>1), returns one value per dimension.

    • if binary_pred_anomalies is a sequence of univariate series, returns one value per series

  • Sequence[Sequence[float]] if binary_pred_anomalies is a sequence of multivariate series. Outer Sequence is over the sequence input, and the inner Sequence is over the dimensions of each element in the sequence input.

Return type

Union[float, Sequence[float], Sequence[Sequence[float]]]

darts.ad.utils.eval_accuracy_from_scores(actual_anomalies, anomaly_score, window=1, metric='AUC_ROC')[source]

Scores the results against true anomalies.

actual_anomalies and anomaly_score must have the same shape. actual_anomalies must be binary and have values belonging to the two classes (0 and 1).

If one series is given for actual_anomalies and anomaly_score contains more than one series, the function will consider actual_anomalies as the ground truth anomalies for all scores in anomaly_score.

Parameters

  • actual_anomalies (Union[TimeSeries, Sequence[TimeSeries]]) – The ground truth of the anomalies (1 if it is an anomaly and 0 if not).

  • anomaly_score (Union[TimeSeries, Sequence[TimeSeries]]) – Series indicating how anomoulous each window of size w is.

  • window (Union[int, Sequence[int]]) – Integer value indicating the number of past samples each point represents in the anomaly_score. The parameter will be used by the function _window_adjustment_anomalies() to transform actual_anomalies. If a list is given. the length must match the number of series in anomaly_score and actual_anomalies. If only one window is given, the value will be used for every series in anomaly_score and actual_anomalies.

  • metric (str) – Optionally, Scoring function to use. Must be one of “AUC_ROC” and “AUC_PR”. Default: “AUC_ROC”

Returns

Score of the anomalies score prediction

  • float if anomaly_score is a univariate series (dimension=1).

  • Sequence[float]

    • if anomaly_score is a multivariate series (dimension>1), returns one value per dimension.

    • if anomaly_score is a sequence of univariate series, returns one value per series

  • Sequence[Sequence[float]] if anomaly_score is a sequence of multivariate series. Outer Sequence is over the sequence input, and the inner Sequence is over the dimensions of each element in the sequence input.

Return type

Union[float, Sequence[float], Sequence[Sequence[float]]]

darts.ad.utils.show_anomalies_from_scores(series, model_output=None, anomaly_scores=None, window=1, names_of_scorers=None, actual_anomalies=None, title=None, metric=None)[source]

Plot the results generated by an anomaly model.

The plot will be composed of the following:

  • the series itself with the output of the model (if given)

  • the anomaly score of each scorer. The scorer with different windows will be separated.

  • the actual anomalies, if given.

If model_output is stochastic (i.e., if it has multiple samples), the function will plot:

  • the mean per timestamp

  • the quantile 0.95 for an upper bound

  • the quantile 0.05 for a lower bound

Possible to:

  • add a title to the figure with the parameter title

  • give personalized names for the scorers with names_of_scorers

  • show the results of a metric for each anomaly score (AUC_ROC or AUC_PR), if the actual anomalies is given

Parameters

  • series (TimeSeries) – The series to visualize anomalies from.

  • model_output (Optional[TimeSeries]) – Output of the model given as input the series (can be stochastic).

  • anomaly_scores (Union[TimeSeries, Sequence[TimeSeries], None]) – Output of the scorers given the output of the model and the series.

  • window (Union[int, Sequence[int]]) – Window parameter for each anomaly scores. Default: 1. If a list of anomaly scores is given, the same default window will be used for every score.

  • names_of_scorers (Union[str, Sequence[str], None]) – Name of the scores. Must be a list of length equal to the number of scorers in the anomaly_model.

  • actual_anomalies (Optional[TimeSeries]) – The ground truth of the anomalies (1 if it is an anomaly and 0 if not)

  • title (Optional[str]) – Title of the figure

  • metric (Optional[str]) – Optionally, Scoring function to use. Must be one of “AUC_ROC” and “AUC_PR”. Default: “AUC_ROC”

previous

WassersteinScorer

next

Data Processing