BaSC is an advanced Bayesian source finding library that uses a likelihood model that is mathematically proven in terms of visibilities but which can operate using only a dirty map; thus gaining accuracy without large computational expense.
Source finding that uses CLEANed maps is subject to the loss of information necessarily caused by that algorithm. In fact, no matter the quality of such algorithms, the best that they can hope to recover is the CLEAN model – not the actual sources.
BaSC works on dirty maps, which still have the messier PSF that is produced by interferometers but do not suffer a loss of information. We apply an MCMC method alongside a likelihood function on the map that is proven equivalent to a (much more expensive) function for the visibilities.
The result is more accurate source positions and fluxes, better ability to discriminate sources near the level of instrument resolution, and a more transparent provenance for the source list produced.
BaSC is provided as a Python 3 library. Download from the Git repository above and, for the case of Linux type ‘make’ or for Mac type ‘make mac’. Windows is not natively supported at this time. A user manual is included in the repository.
This code will continue development at Cambridge once ASTERICS is concluded. Focus will be on included more functionality within BaSC – at the moment generation of the dirty maps has to be done externally with programs such as CASA, but there is an advantage of being able to process the visibilities directly within BaSC before applying the source finder.
Optimisation of the MCMC process is also being looked at; at present, the source finder can take a long time to run in the case of many (>30) sources due to degeneracies in the parameter space. We aim to improve this.
Wide field images can cause issues in BaSC due to the change in the PSF across the field. There are a number of ways to address this which are being investigated
This time period was concerned with the development and testing of BaSC. The software is now in a useable form, publicly available on github. We have produced a paper (Hague et al. 2019) that outlines the results of the tests we have done, and confirms that the software is superior at source discrimination tasks than the usual pathway for radio astronomy of CLEAN/SExtractor and approaches the mathematically optimal resolution limit.
At the beginning of this project, the challenge was correctly packaging an older MCMC code with a python wrapper and a new likelihood function. There was also the challenge of clustering the output.
We were able to calculate the optimal performance for a constrained task (discriminating between two nearby points) and then construct appropriate testing sets. The design of the experiment was critical to confirming that BaSC did indeed work as expected.
The main effort was in programming BaSC itself, and creating realistic test observations that permitted the experiment .Production of the paper and the meeting the referees requirements also took up time.
More features for analysis of radio interferometry images will be included in the package
Haoyang Ye and Peter Hague