Skip to content

Latest commit

 

History

History
164 lines (112 loc) · 4.57 KB

anomaly-types.md

File metadata and controls

164 lines (112 loc) · 4.57 KB
Raw

Anomaly Types

The generator comes with the following anomaly types in gutenTAG/anomalies/types:

  • amplitude
  • extremum
  • frequency
  • mean
  • pattern
  • pattern-shift
  • platform
  • trend
  • variance
  • mode-correlation

General Anomaly Properties

Name Type Description
kind Enum['amplitude','extremum','frequency','mean','pattern','pattern_shift','platform','trend','variance'] Kind of anomaly (see below)
position Enum['beginning','middle','end'] Position of anomaly
length Int Length of anomaly
channel Int ID of channel on which anomaly should be applied

Amplitude

amplitude

Parameters

Name Type Description
amplitude_factor Float Change factor of amplitude

Extremum

extremum

Parameters

Name Type Description
min Bool Whether it's a minimum or maximum
local Bool Whether it's local or global
context_window Int [Only local] how many points to the left and right are used to calculate extremum

Frequency

frequency

Parameters

Name Type Description
frequency_factor Float Relative change of frequency

Mean

mean

Parameters

Name Type Description
offset Float Value to shift time series on Y-axis

Pattern

pattern pattern

Parameters

Name Type Description
sinusoid_k Float [Only for sine and cosine] Ramming factor for changing sine waves.
square_duty Float [Only for square] New duty of the square wave.
sawtooth_width Float [Only for sawtooth] New width
cbf_pattern_factor Float [Only for cylinder_bell_funnel] Pattern variance factor for change in CBF wave.

Only one of the parameters above is necessary. sinusoid_k works with sine BOs, while cbf_pattern_factor works with cylinder-bell-funnel BOs.

Pattern Shift

pattern-shift

Parameters

Name Type Description
shift_by Int Size of the shift length to the right. Can be negative for shift to the left.
transition_window Int Number of points to the left and right used for transition.

Platform

platform

Parameters

Name Type Description
value Float Value of the platform on Y-axis

Trend

trend

A trend anomaly can have any form of the base oscillations. Use the same parameters as for the base oscillations.

Variance

variance

Parameters

Name Type Description
variance Float Value of the new variance

Mode Correlation

mode-correlation

Takes 1-d subsequence of arbitrary length and changes the mode to the opposite.

Parameters

None

Adding a new Anomaly Type

  1. create a new Enum type for AnomalyKind and adapt the generate method
  2. [RECOMMENDED] create a new anomaly type class under gutenTAG/anomalies/types
    1. the new class should inherit from gutenTAG.anomalies.BaseAnomaly

Creeping Anomaly

Some anomaly types (amplitude, mean, and variance) allow for a transition into the anomaly - a creeping. Therefore, the additional parameter creeping-length for an anomaly is introduced. It takes creeping-length points from the overall length anomaly points and creates a linear transition into the length - creeping-length anomaly.

Example

timeseries:
  - name: variance
    length: 1000
    base-oscillations:
      - kind: sine
        frequency: 2
        variance: 0.1
    anomalies:
      - exact-position: 460
        length: 540
        creeping-length: 500
        channel: 0
        kinds:
          - kind: variance
            variance: 1.0

example-anomaly-creeping.png