Energy and polarization based on-line interference mitigation in radio astronomy. Methods are described in this paper.
Python code for on-line RFI mitigation using spectral kurtosis and polarization alignment of radio interferometric data.
Example:
flagpol.py --MS data.MS --finite_prec --time_window_size 10 --freq_window_size 2
will use data.MS as the input data and perform flagging. Using --finite_prec will turn on finite precision emulation, hence slower, so just to test the flagging algorithms themselves, do not enable this option.
Python code for training a reinforcement learning (RL) agent for optimizing precision of arithmetic operations used in the flagging algorithm.
Example:
main_sac.py --episodes 100000 --seed 3333
will train an RL model to optimize the precision of the computing routines (cuda, 32 bit or 16 bit) using the soft actor-critic algorithm. After an ensemble of such models are trained (with different random --seed), you can store each model in directorites like mydir/run1/, mydir/run2/, mydir/run3/ and so on. Thereafter, run the ensemble evaluation as
eval_model.py --episodes 100000 --steps 100 --models 4 --path mydir
Python code for simulating realistic data with known RFI, and performing RFI mitigation. Thereafter, calculating false alarm and missed detection probabilities.
pytorch, numpy, scipy, python-casacore, gymnasium, matplotlib
wo 2 apr 2025 10:20:45 CEST