Joint Image Framework (JIF) for probabilistic modeling of astronomical images of stars and galaxies in optical wavelengths.
How do we optimally combine images of galaxies seen from space and ground? The different PSFs, wavelength coverage, and pixel sizes can lead to biases in inferred galaxy properties unless included in a joint model of all images of the same source. If sources are blended together in any observations, the need for joint modeling becomes even more acute.
This package embeds GalSim image models of galaxies and stars into a Markov Chain Monte Carlo (MCMC) framework for probabilistic forward modeling of images. The primary module is jiffy/roaster.py, which defines the image model likelihood. Most parameters to roaster can be specified in configuration files as in config/jiffy.yaml.
To create a conda environment named "jiftutorial" and install the minimum necessary packages:
conda create -n jiftutorial python=3.8.12 numpy tqdm h5py yaml matplotlib jupyter astropy pandas scipy
conda activate jiftutorial
conda install -c conda-forge galsim emcee scikit-learn
python -m pip install -U corner
In addition, this package (JIF) as well as the footprints package need to be cloned and installed from their respective repos, as follows:
git clone [email protected]:mdschneider/footprints.git
git clone [email protected]:mdschneider/JIF.git
cd footprints
python setup.py install
cd ../JIF
python setup.py install
Everything in this project is in python (not including our external dependencies). Install with,
python setup.py install
or
python setup.py develop
to install while working on the package.
Install python requirements available with PIP via:
pip install -r requirements.txt
- Our image models are built around the GalSim image simulation framework.
- For parameter inference, we use emcee.
- The sources (and source groups) we extract from raw imaging are stored in HDF5 file formats, with a custom grouping.
- For part of the results visualization we use corner.
Branch 0.1 contains the specific code used for the first arXiv draft of "Markov Chain Monte Carlo for Bayesian Parametric Galaxy Modeling in LSST", submitted 19 September 2023.