GalacticFlow

GalacticFlow applies machine learning to astrophysics. It uses conditional normalizing flows to learn a generalized representation of galaxies. It learns the (extended) distribution function of galactic (stellar) data conditioned on global galactic parameters, such as, e.g., total stellar mass. Galactic flow then provides a compact generative model that allows sampling a galaxy at any desired parameters with any desired number of stars, as well as exactly evaluating the distribution function, essentially interpolating and generalizing in galactic parameter space.

<img src="https://github.com/luwo9/GalacticFlow/assets/126659866/e3a9c26e-306c-4a0d-8981-e939fabcc127" alt="Galactic Flow scheme" width="700">

User guide

While it is possible to work on a very low level with maximum flexibility, see Base_Processor_Workflow.ipynb, its strongly recommended to use the user friendly API. The usage of the API is documented in API_Workflow.ipynb, it allows loading pretrained models, defining and taining your own ones, sampling a glaxy, as well as evaluating its pdf with minimum effort and code. It implements the low level workflow under the hood, so it might still be insightful to read Base_Processor_Workflow.ipynb.

Reproducing NeurIPS paper results

reproduce.ipynb also provides the direct teps to reproduce the results in the paper. The packages used in our python (3.8.10) environment can be found in requirements.txt (not all packages are necessary to run GalacticFlow). The data used, as well as the pre trained models can be found at Zenodo for download. The data must be contained in the all_sims folder in your current directory for the pre trained models to work, you can put the models in any folder and specify the right path whenever needed.

Reusing the code

Adapting_GF.ipynb provides more details on how to adapt the code to your own data and needs.