GaussianScorer

Univariate Gaussian anomaly detector.

This is a supervised anomaly detector. It fits a Gaussian distribution to the target values. The anomaly score is then computed as so:

\[score = 2 \mid CDF(y) - 0.5 \mid\]

This makes it so that the anomaly score is between 0 and 1.

Parameters

• window_size

  Default → None

  Set this to fit the Gaussian distribution over a window of recent values.

• grace_period

  Default → 100

  Number of samples before which a 0 is always returned. This is handy because the Gaussian distribution needs time to stabilize, and will likely produce overly high anomaly score for the first samples.

Examples

import random
from river import anomaly

rng = random.Random(42)
detector = anomaly.GaussianScorer()

for y in (rng.gauss(0, 1) for _ in range(100)):
    detector = detector.learn_one(None, y)

detector.score_one(None, -3)

0.999477...

detector.score_one(None, 3)

0.999153...

detector.score_one(None, 0)

0.052665...

detector.score_one(None, 0.5)

0.383717...

Methods

learn_one

Update the model.

Parameters

• x — 'dict'
• y — 'base.typing.Target'

Returns

SupervisedAnomalyDetector: self

score_one

Return an outlier score.

A high score is indicative of an anomaly. A low score corresponds a normal observation.

Parameters

• x — 'dict'
• y — 'base.typing.Target'

Returns

float: An anomaly score. A high score is indicative of an anomaly. A low score corresponds a