Estimating Dark Matter Halo Masses with Machine Learning

This repository contains the code and data used in the paper "Estimating Dark Matter Halo Masses in Simulated Galaxy Clusters with Graph Neural Networks" (Garuda et al. 2024).

Data

We use TNG-Cluster snapshot 99 $z=0$ from the TNG Cluster simulations. This data will later be released by the IllustrisTNG team.

Requirements

This code requires you to download pytorch and torch-geometric; you can install them by checking the PyTorch Geometric website.

But, you can also use the requirements.txt file to install all the dependencies. You can do this by running the following command:

pip install -r requirements.txt

Usage

All the source required to train the model and make your datasets from the raw data is provided in the src directory. Best method would be to use a Weights and Biases account to log all the results. You can create a free account here.

Easiest way would be to create the datasets using data_load.py and then changing your file paths in wandb_run_advgnn.py and running the following command:

wandb sweep sweep.yaml 

This will run the sweep and log all the results to your Weights and Biases account.

Citation

TODO