Department of Mathematics and Statistics, Thompson Rivers University, Kamloops, BC, Canada V2C0C8;
Department of Mathematics and Statistics, York University, Toronto, ON, Canada M3J1P3;
Proc Natl Acad Sci U S A. 2018 Jun 5;115(23):5914-5919. doi: 10.1073/pnas.1804649115. Epub 2018 May 21.
The change-point detection has been carried out in terms of the Euclidean minimum spanning tree (MST) and shortest Hamiltonian path (SHP), with successful applications in the determination of authorship of a classic novel, the detection of change in a network over time, the detection of cell divisions, etc. However, these Euclidean graph-based tests may fail if a dataset contains random interferences. To solve this problem, we present a powerful non-Euclidean SHP-based test, which is consistent and distribution-free. The simulation shows that the test is more powerful than both Euclidean MST- and SHP-based tests and the non-Euclidean MST-based test. Its applicability in detecting both landing and departure times in video data of bees' flower visits is illustrated.
已经针对欧几里得最小生成树 (MST) 和最短哈密顿路径 (SHP) 进行了变点检测，在确定经典小说的作者、检测网络随时间的变化、检测细胞分裂等方面都取得了成功的应用。然而，如果数据集包含随机干扰，这些基于欧几里得图的测试可能会失败。为了解决这个问题，我们提出了一种强大的基于非欧几里得 SHP 的测试方法，该方法具有一致性和分布自由性。模拟表明，该测试比基于欧几里得 MST 和 SHP 的测试以及基于非欧几里得 MST 的测试都更有效。它在检测蜜蜂访问花朵的视频数据中的着陆和离开时间方面的适用性也得到了说明。
