Repository for CNN models as presented in A Robust and Efficient Deep Learning Method for Dynamical Mass Measurements of Galaxy Clusters. This repository is meant to serve as an example of the code used to perform this deep learning analysis, not as a fully-developed utility for public use.
The mock cluster catalog generation script is make_mocks.py. It uses halo data from the MultiDark Planck 2 simulation Rockstar catalog and a galaxy catalog generated using UniverseMachine. The generated catalogs are stored as Catalog objects, detailed in catalog.py. For a full previously-generated mock catalog, reach out to the corresponding author at [email protected].
A brief tutorial on dataset preprocessing and model fitting is given in <tutorial.ipynb>. Data processing is handled by the HaloCNNDataManager class in data.py. Models are represented as the BaseHaloCNNRegressor class in model.py.