Skip to content

Latest commit

 

History

History
102 lines (73 loc) · 4.85 KB

index.md

File metadata and controls

102 lines (73 loc) · 4.85 KB

Introduction

The GutenTAG is a good timeseries anomaly generator. As a user, you can choose between several Base Oscillations and Anomaly Types which are applied on the base oscillations. However, not all anomalies can be applied on all base oscillations. An overview can be found here.

Installation

You can install GutenTAG from the official Python Package Index or from source.

  • Installation from PyPI:

    pip install timeeval-gutenTAG

  • Installation from source:

    git clone [email protected]:TimeEval/GutenTAG.git

cd GutenTAG

# (optionally) create a new conda environment with Python 3
conda create -n gutentag python=3.8
conda activate gutentag

# install dependencies
pip install -r requirements.txt

Test the installation with python -m gutenTAG and you should see the greeting and usage instructions:

$ python -m gutenTAG


                      Welcome to

       _____       _          _______       _____ _
      / ____|     | |        |__   __|/\   / ____| |
     | |  __ _   _| |_ ___ _ __ | |  /  \ | |  __| |
     | | |_ | | | | __/ _ \ '_ \| | / /\ \| | |_ | |
     | |__| | |_| | ||  __/ | | | |/ ____ \ |__| |_|
      \_____|\__,_|\__\___|_| |_|_/_/    \_\_____(_)

                     Version 1.2.0

"Good day!" wishes your friendly Timeseries Anomaly Generator.



usage: __main__.py [-h] [--version] --config-yaml CONFIG_YAML
                   [--output-dir OUTPUT_DIR] [--plot] [--no-save]
                   [--seed SEED] [--addons [ADDONS [ADDONS ...]]]
                   [--n_jobs N_JOBS] [--only ONLY]
__main__.py: error: the following arguments are required: --config-yaml

Go to the usage-page for further instructions on how to call GutenTAG.

Structure

GutenTag generates time series based on two fundamental building blocks: base oscillations and anomalies. Each time series has the following properties:

  • optional name
  • length
  • a list of base oscillations defining the different channels
  • a list of anomalies (at different positions in the time series)
  • (some additional optional configuration options)

Internal Process

  1. We first generate the base oscillation.
  2. Then, we inject the anomalies at the specific locations.
    1. The location of an anomaly is defined by the given position (in time)
      1. A position can be an exact time stamp or index in the time series
      2. or an approximate region like beginning, middle, or end. Here, GutenTAG will randomly choose a cycle in the corresponding region and will use its start as position.
    2. and the channel defined by the channel ID.
  3. Afterwards, the variations like noise, trend, and offset are applied to the time series.

Combinations

The following table shows which anomalies and base oscillations can be combined and which combinations GutenTAG does not support.

  • x = Combination allowed
  • - = Combination not allowed
Sine Cosine Square Random Walk CBF ECG Polynomial Random Mode Jump Formula Sawtooth Dirichlet MLS Custom Input
amplitude x x x x x x - - - x x x x
extremum x x x x x x x - x x x x x
frequency x x x - - x - - - x x - -
mean x x x x x x x - x x x x x
pattern x x x - x x - - - x x x -
pattern_shift x x x - - x - - - x x - -
platform x x x x x x x - x x x x x
trend x x x x x x x - x x x x x
variance x x x x x x x - x x x x x
mode_correlation - - - - - - - x - - - - -