This is a component of ML-PPA (Machine Learning-based Pipeline for Pulsar Analysis), a framework for extracting pulsar signals in data streams from radio astronomical antennas.
The main tool of the framework.
Release Notes
This is the first internal release. Assorted related materials, including Jupyter notebooks with use examples, are available at GitLab or will be added soon. For a wider context and theory behind the whole ML-PPA (including detailed explanations with regard to the status of each component) one should refer to the paper. The main functionality of this component is assigning labels to 2D time-frequency “timeframes”. Each timeframe can be classified into one of the 4 main categories: pulse, two kinds of interference, or empty, i.e. just noise. To do this the tool must be first trained using either real data or that generated by either physics-based DT (PulsarDT) or empirical DT (PulsarRFI_Gen).
Users are also supplied with Jupyter notebooks that illustrate the use of this component.
Future Plans
This tool accomplishes its primary purpose. However, its performance when dealing with data with a low signal-to-noise ratio still needs to be improved, and that is going to be the focus of the development efforts in the near future. In particular, the ML approach is being changed (distributed training: HeAT, Horovod) and noise is to be handled more efficiently. And in a more distant future the plans are for it to be able to classify much more than just 4 basic categories of data, ideally to detect physical properties of the signal and interference components.