A useful set of classes and utilities to produce self-organising maps with PINK, annotate these neurons and perform object collation with the outputs.
The process has three main steps:
- create a SOM using
PINKthat accurately describes the predominate features within an image set, - annotate the SOM in following some user defined classification scheme, and
- transfer these labels from the SOM and it's neurons to real on-sky catalogue data by using the mapping and spatial transform solutions returned by
PINK.
A more thorough description is provided by Galvin et al. (submitted), who use image data at radio and infrared wavelengths to demonstrate multi-wavelength host identification is possible within an unsupervised framework.
Throughout we use PINK v2 while developing this module. Note that the binary files produced by PINK and the corresponding classes are not directly interchangeable between PINK version 1 and 2 releases.
- numpy
- scipy
- astropy
- scikit-image
- tqdm
Collaborators and pull requests are welcome. Please do not hesitate to ask questions or get involved.
Throughout the code base type annotations, doc-strings and the black code formatter have been used, with mypy as a linter.