A 2D Design Space defined with non-linear equations using different sampling methods with different number of data points

A 2D design space with two parameters were created using different sampling methods: grid, Latin hypercube sampling (LHS), random, and antithetic version of the last two. The number of sample points to cover the study space are: 100, 225, 625, 1225, and 2500. The lower values for both parameters equal to 0.2 and upper values equal to 1. The design space is based on the geometry characterised by non-linear equations, and non-convexity. The synthetic tabular datasets contain two parameters and consider a binary classification problem, where points are “Good” denoted with “1” if they are in the interior of the design space and “Bad” denoted with “0” if they are not. The datasets were used to extract two extra datasets to train, evaluate, and compare classification models coupled with active learning strategies. The two extra datasets extracted from the datasets containing the values of parameters and the target associated are: (i) the indexes of the initial labelled samples and (ii) the indexes of the initial training samples.

本研究采用四种采样策略构建含两个参数的二维设计空间，分别为网格采样、拉丁超立方采样（Latin Hypercube Sampling, LHS）、随机采样，以及后两种采样方法的对偶版本。用于覆盖该研究空间的采样点数目依次为100、225、625、1225与2500。两个参数的取值下限均为0.2，上限均为1。该设计空间基于由非线性方程表征且具有非凸特性的几何结构。所生成的合成表格数据集包含两个参数，并对应一项二分类任务：若采样点位于设计空间内部，则标记为“Good”，对应标签“1”；若位于设计空间外部，则标记为“Bad”，对应标签“0”。上述数据集被用于衍生两个额外数据集，以训练、评估并对比结合主动学习策略的分类模型。从包含参数取值与对应标签的原始数据集中衍生的两个额外数据集分别为：(i) 初始带标签样本的索引集；(ii) 初始训练样本的索引集。