知识卡片
模型验收阈值包含错误代价取舍
F分数等指标可能压缩不同错误和相关角色的利益冲突,阈值还受业务代理关系与随机波动影响。需先协商取舍、逐步设定合理改进,并用统计思维避免偶然通过或失败,不能只填一个理想分数。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 产品与交付实践
原书出处 · 1 处
- · 查看原书依据
Testable principle, in particular, presents unique challenges in ML, especially when expressed as a threshold for a given measure (let’s take F-score as an example). First, the chosen measure may involve reconciling objectives in conflict or on a Pareto front (false positives and false negatives in the case of F-score) and typically has a political element, reflecting the influence of various stakeholders and who stands
参考来源
- 位置:《Effective Machine Learning Teams (David Tan, Ada Leung, David Colls)》第02章《产品与交付实践》(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/ch02.html)
- 结论依据:摘录中关于模型验收阈值包含错误代价取舍的定义、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:Testable principle, in particular, presents unique challenges in ML, especially when expressed as a threshold for a given measure (let’s take F-score as an example). First, the chosen measure may involve reconciling objectives in conflict or on a Pareto front (false positives and false negatives in the case of F-score) and typically has a political element, reflecting the influence of various stakeholders and who stands