Galaxy Clustering Dataset
收藏资源简介:
Galaxy Clustering Dataset是由麻省理工学院的研究团队从Quijote N-body模拟中提取的,用于测试点云处理算法的数据集。该数据集包含12,384个模拟的星系位置和属性,每个模拟包含5000个最重的暗物质晕的3D坐标。数据集的创建过程涉及复杂的模拟和处理,旨在捕捉宇宙的大规模结构和多尺度信息。该数据集主要应用于宇宙学领域,旨在解决从星系分布中提取有意义信息的问题,特别是关于暗物质分布和宇宙膨胀历史的推断。
The Galaxy Clustering Dataset was extracted from the Quijote N-body simulations by a research team at the Massachusetts Institute of Technology (MIT), serving as a benchmark dataset for testing point cloud processing algorithms. This dataset encompasses 12,384 sets of simulated galaxy positions and their associated properties, with each individual simulation containing the 3D coordinates of the 5000 most massive dark matter halos. The development of this dataset involves complex simulation and processing procedures, with the goal of capturing the large-scale cosmic structure and multi-scale cosmological information. Primarily utilized in the field of cosmology, this dataset is intended to extract meaningful information from galaxy distributions, specifically for inferring dark matter distributions and the historical expansion of the universe.
$E(3)$ Equivariant Graph Neural Networks in Jax
数据集概述
该数据集包含用于训练和测试的宇宙学基准数据,以TFRecord格式存储。数据集可通过Zenodo下载,DOI为10.5281/zenodo.11479419。
数据集下载
数据集下载脚本位于benchmarks/galaxies/download_tfrecords.sh,运行该脚本可将数据集下载至benchmarks/galaxies/quijote_records目录。
数据集使用
- 图级别任务:运行
python benchmarks/galaxies/train_cosmology.py。 - 节点级别任务:运行
python benchmarks/galaxies/train_velocities.py。
数据集引用
引用信息详见CITATION.cff。




