← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8762.c · 生成模型:GPT-6

聚合测试不能刻画全部组合

理解等级:中级

即使总体和单维切片合格,多条件组合仍可能触发失败。行为测试可针对这些条件设计样本,真实分布变化又需新评价数据;不能用更多同分布指标替代未覆盖场景。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      (i.e., exercise the model based on all scenarios present in the available data ), it doesn’t allow you to characterize pockets of behavior within segments of the data. Even in the case of stratified metrics tests, there may be combinatorial effects in multiple dimensions that can confound a model, but it can be hard to specify, characterize, and discover such scenarios in a metrics test.

    参考来源

    - 位置:《Effective Machine Learning Teams (David Tan, Ada Leung, David Colls)》第06章《模型自动化测试》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Effective Machine Learning Teams (David Tan, Ada Leung, David Colls)/_epub-src/OEBPS/ch06.html) - 结论依据:摘录中关于聚合测试不能刻画全部组合的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:(i.e., exercise the model based on all scenarios present in the available data ), it doesn’t allow you to characterize pockets of behavior within segments of the data. Even in the case of stratified metrics tests, there may be combinatorial effects in multiple dimensions that can confound a model, but it can be hard to specify, characterize, and discover such scenarios in a metrics test.