This repository provides an implementation of the Deep Ensemble Anomaly Detection for Time Series (DEAN-TS) method that I developed as part of my master thesis. It is based on Deep Ensemble Anomaly Detection (DEAN) and, as the name suggests, applies its concepts to time series.
- Install requirements according to
requirements.txt.
- Specify parameterization of DEAN-TS in
config/configuration.yaml.
- Specify the execution details and apply DEAN-TS by customising
src/main.pyand then running it, e.g. viapython main.py.
A substantial part of the evaluation in my master thesis is performed with the benchmark tool TimeEval,
among others on synthetic datasets generated with the generation tool GutenTAG by the same developers.
In particular, DEAN-TS expects input formatting as defined there.
Some of the synthetic datasets used are included in datasets/ecg for easy experimentation with DEAN-TS.
Information on the integration of DEAN-TS into TimeEval can be found here.