Scripts and data for PRADS publication that varies the acquisition function of the Adaptive Screening method

This repository can be used to re-generate the results in the paper referenced below. Abstract of the paper: "The Adaptive Screening extreme value prediction method defined in \cite{VS2025} was applied to three cases, predicting most probable maximum values in a given duration. We used either Gaussian Process Regression (GPR) or its multi-fidelity form (MF-GPR) for the regression, and compared seven acquisition functions for the adaptive sampling. The results were judged based on several performance metrics, of which the required number of HF samples for convergence is the most important one. The results of this exercise show that the best option is to use Adaptive Screening with MF-GPR and an acquisition function that balances obtaining new samples around the probability of interest and the tail of the distribution. If the use of MF-GPR is not feasible for any reason, it should be replaced with GPR (or Generalised Pareto fitting) combined with an acquisition function that selects new samples based on the largest probability gap from the existing samples."Van Essen, S.M. and Seyffert, H.C. (2025). Designing ships for extreme non-linear responses - the role of the acquisition function in the Adaptive Screening method. 16th Practical Design of Ships and other Floating Structures (PRADS) Conference, 19-23 Oct, Ann Arbor, USA.

本仓库可用于复现下述参考文献中论文的研究结果。论文摘要如下："本文将文献cite{VS2025}中提出的自适应筛选极值预测方法（Adaptive Screening extreme value prediction method）应用于三个案例，以预测给定时长内最具可能性的最大值。实验分别采用高斯过程回归（Gaussian Process Regression, GPR）及其多保真度形式（MF-GPR）完成回归任务，并对比了七种用于自适应采样的采集函数。模型性能通过多项评估指标进行评判，其中达到收敛所需的高保真（HF）样本数量为最核心的评价指标。本研究结果表明，最优方案为结合MF-GPR的自适应筛选方法，搭配能够兼顾目标概率区域与分布尾部区域采样的采集函数。若因任何原因无法使用MF-GPR，则可替换为GPR（或广义帕累托拟合），并采用基于现有样本最大概率间隙选取新样本的采集函数。"参考文献：Van Essen, S.M. 与 Seyffert, H.C. (2025) 所著《面向极端非线性响应的船舶设计——采集函数在自适应筛选方法中的作用》，收录于第16届国际船舶及其他浮动结构实用设计（Practical Design of Ships and other Floating Structures, PRADS）会议，该会议于2025年10月19日至23日在美国安娜堡举办。