lamm-mit/BeamRL-EvalData-v2
收藏资源简介:
BeamRL-EvalData-v2是一个扩展的评估数据集,专为BeamPERL项目设计,包含123个符号梁力学问题,旨在计算支座反力。数据集中每个问题的真实答案以负载符号P的带符号系数形式给出(例如0.222222P)。数据集涵盖多个类别:id(单点载荷,支座在梁端)、ood_loads(2-4个点载荷,支座在梁端)、ood_supports(不同支座位置,1-3个点载荷)、ood_dist_left/ood_dist_middle/ood_dist_right(分布载荷分别锚定在铰支座、完全内部或滚支座上)、ood_length(梁长度变化为7L、11L、13L)和ood_moment(纯应用力偶,无点载荷)。原始24样本评估集作为子集包含在内。所有配置均使用修改版的SymBeam符号求解,并通过力/力矩平衡验证;自然语言问题变体由LLM生成,每个样本有4个查询,评估工具使用query[0]。数据集由BeamPERL仓库中的DataGen/generate_eval_v2.py脚本生成。
BeamRL-EvalData-v2 is an expanded evaluation dataset for BeamPERL, consisting of 123 symbolic beam-mechanics problems that ask for the reaction forces at the supports, with ground-truth answers as signed coefficients of the load symbol P (e.g., 0.222222P). The dataset includes categories such as id (single point load, supports at beam ends), ood_loads (2–4 point loads, supports at beam ends), ood_supports (varying support positions, 1–3 point loads), ood_dist_left/ood_dist_middle/ood_dist_right (distributed load anchored to the pin/fully interior/anchored to the roller), ood_length (beam length varied as 7L, 11L, 13L), and ood_moment (pure applied couple, no point load). The original 24-sample evaluation set is included as a subset. All configurations are solved symbolically with a modified SymBeam and verified for force/moment balance; natural-language question variants (4 per sample, with the evaluation harness consuming query[0]) are generated with an LLM. The dataset is produced by DataGen/generate_eval_v2.py in the BeamPERL repository.




