StreamAD Process

Post process

ZScoreThresholder

class streamad.process.ZScoreThresholder(detector, sigma=3, is_global=True, window_len=100)

Bases: object

__init__(detector, sigma=3, is_global=True, window_len=100)

A thresholder which can filter out outliers using z-score, and normalize the anomaly scores into [0,1].

Parameters

• detector (BaseDetector) – A detector that must be a child class of BaseDetector.

• sigma (int, optional) – Zscore threshold, we regard the scores out of sigma as anomalies. Defaults to 3.

• is_global (bool, optional) – Method to record, a global way or a rolling window way. Defaults to True.

• window_len (int, optional) – The length of rolling window, ignore this when is_global=True. Defaults to 100.

---

TDigestThresholder

class streamad.process.TDigestThresholder(detector, percentile_up=95, percentile_down=5, is_global=True, window_len=100)

Bases: object

__init__(detector, percentile_up=95, percentile_down=5, is_global=True, window_len=100)

A thresholder which can filter out outliers using t-digest, and normalize the anomaly scores into [0,1] [Dunning, 2021].

Parameters

• detector (BaseDetector) – A detector that must be a child class of BaseDetector.

• percentile_up (float, optional) – We regard the scores above percentile_up as anomalies. Defaults to 95.

• percentile_down (float, optional) – We regard the scores below percentile_down as anomalies. Defaults to 5.

• is_global (bool, optional) – Method to record, a global way or a rolling window way. Defaults to True.

• window_len (int, optional) – The length of rolling window, ignore this when is_global=True. Defaults to 100.