遇见数据集

ClarusC64/DF-delayed-feedback-probe-v1

收藏
Hugging Face2026-04-08 更新2026-04-12 收录
官方服务:

资源简介:

--- language: en license: mit task_categories: - tabular-classification tags: - stability-intelligence - instability-geometry - cascade-collapse - clarus size_categories: - n<1K pretty_name: Geometry - Cascade Collapse Detection v1 --- # What this dataset tests This dataset tests whether a model can detect the structural signature of cascade collapse from a compact state-space snapshot. # Core instability geometry Cascade collapse occurs when pressure rises inside a tightly coupled system, buffers are too weak to absorb the load, and drift is already oriented toward instability. Local strain then propagates across subsystems rather than remaining contained. # Prediction target `label_cascade_forming` - `1` = a cascade pattern is forming - `0` = no cascade pattern is forming # Row structure Each row is one scenario. Columns: - `scenario_id` - `pressure` - `buffer_capacity` - `coupling_strength` - `intervention_lag` - `drift_gradient` - `label_cascade_forming` (train only) # Signal definitions - `pressure`: current system load or strain - `buffer_capacity`: available capacity to absorb further strain - `coupling_strength`: how strongly subsystem stress propagates - `intervention_lag`: delay before stabilizing action can take effect - `drift_gradient`: direction and intensity of movement toward or away from instability # Files - `data/train.csv` - 10 labeled training rows - `data/tester.csv` - 10 unlabeled evaluation rows - `scorer.py` - minimal binary classification scorer - `README.md` - dataset card # Evaluation Predictions should be submitted as a CSV with: - `scenario_id` - `prediction` where `prediction` is binary `0` or `1`. Example: ```csv scenario_id,prediction cc_test_001,1 cc_test_002,0 Run: python scorer.py predictions.csv ground_truth.csv Why this matters Most benchmarks ask whether a model can classify content. This probe asks whether a model can recognize a failure geometry. That distinction is central to stability intelligence work. License MIT

--- 语言:英语 许可证:MIT 任务类别: - 表格分类(tabular-classification) 标签: - 稳定性智能(stability-intelligence) - 失稳几何(instability-geometry) - 级联失效(cascade collapse) - Clarus(clarus) 样本规模: - n<1K 数据集名称:几何——级联失效检测v1(Geometry - Cascade Collapse Detection v1) --- # 本数据集测试目标 本数据集用于测试模型能否从紧凑的状态空间快照(state-space snapshot)中识别出级联失效的结构特征。 # 核心失稳几何特性 当紧耦合系统内部压力攀升、缓冲能力不足以承载负载,且系统漂移已朝向失稳方向时,便会发生级联失效。此时局部应变会跨子系统(subsystem)扩散,而非被局限在局部范围内。 # 预测目标 `label_cascade_forming` - `1`:表示正形成级联失效模式 - `0`:表示未形成级联失效模式 # 数据行结构 每一行对应一个测试场景。 列信息如下: - `scenario_id`:场景标识符 - `pressure`:系统当前负载或应变 - `buffer_capacity`:吸收额外应变的可用容量 - `coupling_strength`:子系统应力的传递强度 - `intervention_lag`:稳定措施生效前的延迟时长 - `drift_gradient`:系统朝向/背离失稳方向移动的方向与强度 - `label_cascade_forming`(仅训练集包含该列) # 信号定义 - `pressure`:系统当前负载或应变 - `buffer_capacity`:用于吸收额外应变的可用容量 - `coupling_strength`:子系统应力的传递强度 - `intervention_lag`:稳定干预措施生效前的延迟时长 - `drift_gradient`:系统朝向或背离失稳状态移动的方向与强度 # 文件说明 - `data/train.csv`:包含10条带标签的训练样本 - `data/tester.csv`:包含10条无标签的评估样本 - `scorer.py`:轻量级二分类评估脚本 - `README.md`:数据集说明文档 # 评估要求 预测结果需以CSV格式提交,文件需包含以下列: - `scenario_id` - `prediction` 其中`prediction`为二进制标签`0`或`1`。 示例: csv scenario_id,prediction cc_test_001,1 cc_test_002,0 执行评估命令:`python scorer.py predictions.csv ground_truth.csv` # 数据集意义 现有多数基准测试仅考察模型对内容的分类能力,而本数据集旨在测试模型能否识别失效几何特征。这一区别正是稳定性智能研究的核心所在。 # 许可证 MIT

提供机构:
ClarusC64
二维码
社区交流群
二维码
科研交流群
商业服务