Welcome to the ECGAN framework!
The ECGAN framework provides a pipeline to preprocess, train and evaluate PyTorch models trained on time series data. The focus lies on anomaly detection in ECG data: this is highlighted by datasets that are supported out-of-the-box. Generally, the aim is to offer tools to automatically preprocess the data, to provide a unified training and evaluation pipeline and toanalyse data, such as spectral analysis (FFT) and embeddings (UMAP/t-SNE).
The current focus of implemented models lies on ECG generation and its use for the detection of rhythmic or morphologic abnormalities in time series, but several other methods are already supported to serve as baseline detection algorithms. The main goal is to offer reproducible implementations of modern algorithms. This results in a rather flexible framework with simple ways to quickly develop new Training or Anomaly Detection methods. We aim to add several classical models for time series analysis (e.g. autoregressive models) in the near future. In the long term we will try to offer more information and tools for various aspects of ECG data processing - from filtering the raw signal to augmenting datasets.
Please get in touch if you have any suggestions for the framework or find missing or wrong information/implementation details. We aim to provide reproducible state-of-the-art methods, so please feel free to improve our existing models or add new models (see Contributing) for more information.