IPC
收藏资源简介:
IPC数据集是由国际规划竞赛(IPC)中的AI规划任务构建的,旨在为基于图的机器学习方法提供基准测试。该数据集包括两个版本:接地版和提升版,均包含大小不一且分布偏斜的图,为图核和图神经网络等计算图模型提供了重大挑战。数据集中的图是有向的,提升版是无环的,这为有向(无环)结构的专用模型提供了测试机会。此外,图生成器和标签是通过计算机编程实现的,使得数据集可以轻松扩展以满足更大规模的需求。IPC数据集主要应用于图分类、回归及相关任务,旨在解决图结构数据学习中的效率和准确性问题。
The IPC dataset is constructed from AI planning tasks in the International Planning Competition (IPC), serving as a benchmark for graph-based machine learning approaches. This dataset includes two versions: the grounded version and the lifted version, both containing graphs with varying sizes and skewed distributions, which poses significant challenges to computational graph models such as graph kernels and graph neural networks. All graphs in the dataset are directed, while the lifted version is acyclic, offering a testing ground for specialized models tailored to directed (acyclic) graph structures. Furthermore, the graph generators and labels are implemented via computer programming, allowing the dataset to be easily scaled to meet larger-scale requirements. The IPC dataset is predominantly applied to graph classification, regression and related tasks, with the objective of addressing efficiency and accuracy challenges in graph-structured data learning.

- 1IPC: A Benchmark Data Set for Learning with Graph-Structured Data巴塞尔大学 IBM 研究 麻省理工-IBM 沃森人工智能实验室 · 2019年



