jansky-research: A CPU-first, reproducible toolkit for amateur radio-astronomy analyses
A tested, CPU-first Python toolkit for amateur, reproducible radio-astronomy analyses, built on the jansky course library, with optional opt-in GPU (ROCm/CUDA-portable, pure-PyTorch) acceleration for its signal-processing and machine-learning components. It bundles more than forty self-contained public-data research slices, each run on real public data, put through an adversarial science-review gate, and written up honestly as an AASTeX paper. Representative slices, by domain: Fast radio bursts — burst statistics on the CHIME/FRB catalogue (frbstats), recovery of FRB 20180916B's 16.35-day activity period (frbperiod), and a uniform Catalog 2 timing and lensed-delay census (frbwait, frblens); Pulsars — ATNF spectra and the P–Pdot diagram (pulsarspec, ppdot), giant-pulse tests, and glitch waiting-time classification (glitchpop); HI & spectral line — the flat inner Milky Way rotation curve from LAB HI 21 cm data (hi) and an environment-split FASHI HI mass function (fashienv); Solar, heliospheric & planetary radio — type III exciter-speed and beam-tracking analyses (solarbursts, windwaves, swaves, triangulate) and Jovian/Saturnian/ice-giant censuses (junodam, skr, vgpra); Faraday, continuum & SETI — the Galactic RM sky and the first RM dipole test (rmsky, rmdipole), a VLASS variability census recovering FK Comae Berenices (vlass), and a Doppler-drift SETI injection-recovery benchmark with an honest null (driftsearch); GPU / machine learning — a device-portable pure-PyTorch Fast DM Transform and DSP suite (fdmt, torchdsp) and neural simulation-based inference of a radio-emitter population (svsbi). Validations and honest negatives alike; every reported number regenerates from the pipeline. Developed collaboratively with Anthropic's Claude; an AI assistant is not an eligible author and is credited in the acknowledgements only.