Draft repo for the ODSL CMB denoising project.
Follow the maria installation instructions to install maria. Then install this package with:
pip install nifty_maria/ -eThe maria and nifty python packages are added as submodules.
The tutorials folder contains some simple tutorial notebooks for using maria. They may be outdated. Also check: maria/docs/source/tutorials.
Check the transformer_maria folder for a simple first (unsuccessful) transformer-based attempt at reconstructing maria data.
nifty_maria contains code for reconstructing maria data with the jax implementation of nifty. There are step-by-step python notebooks which build up to a simultaneous fit for disentangling astronomical signals from atmosphere contributions:
- nifty_maria/nifty_maria/: jax-based rewrite of maria's map sampling as well as general steering code.
- nifty_maria/map/: Reconstruction of astronomical signals without atmosphere or CMB contributions.
- nifty_maria/atmosphere_mapsampling/: Reconstruction of (static) atmosphere contributions using nifty's 2D correlated field model (i.e. atmosphere "images"). Moderately successful and computationally very heavy!
- nifty_maria/atmosphere_tods/: Reconstruction of 1D atmosphere time-series directly without modelling detector response. More lightweight/successful, but: brings some loss of generality/expressivity since all atmosphere TODs are assumed to be identical, barring pixel fluctuations.
- nifty_maria/simultaneous_fit/: Simultaneous reconstruction of map (image) and atmosphere (time-series). Most successful approach thus far. Better reco than maria baseline.
- nifty_maria/steering_dev/: Test notebooks for testing improved steering notebooks.
- nifty_maria/tests/: Loose collection of small test notebooks.