PointPattern

PointPattern 是由统计力学中的简单点模式构建的图形分类数据集。作者在二维中模拟了三点模式：平衡硬盘 (HD)、泊松点过程和磁盘的随机顺序吸附 (RSA)。 HD 和 Poisson 分布可以看作是描述液体和气体微观结构的简单模型，而 RSA 是一种非平衡随机过程，它在非重叠条件下一个接一个地引入新粒子。 众所周知，这些系统结构不同，但易于模拟，因此提供了可靠且可控的分类任务。对于每个点模式， 粒子被视为节点，随后根据两个粒子是否在阈值距离内绘制边缘。The PointPattern dataset is a graph classification dataset constructed from simple point patterns in statistical mechanics. The authors simulated three types of point patterns in two dimensions: equilibrium hard disks (HD), Poisson point processes, and random sequential adsorption (RSA) of disks. HD and Poisson point distributions can be regarded as simple models describing the microstructures of liquids and gases, while RSA is a non-equilibrium stochastic process that introduces new particles sequentially under non-overlapping constraints. As is widely recognized, these systems exhibit distinct structures yet are straightforward to simulate, thereby providing a reliable and controllable classification task. For each point pattern, particles are treated as nodes, and edges are subsequently established based on whether two particles fall within a specified threshold distance.

搜集汇总

背景与挑战

背景概述

PointPattern是一个图形分类数据集，基于统计力学中的平衡硬盘、泊松点过程和随机顺序吸附三种点模式构建，通过粒子节点和距离阈值生成边缘。该数据集由普林斯顿大学等机构于2020年发布，旨在提供可靠且可控的分类任务。