知识卡片
长训练仍需快速前置检查
当完整训练耗时很长,先以小样本冒烟与快速软件检查捕获明显错误,再启动昂贵训练和完整模型评价。对无需重训的组件则不应借ML之名延迟集成。分层反馈降低等待,但小样本通过不能代替完整质量验收。
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
原书出处 · 1 处
- · 查看原书依据
training smoke tests as described in Chapter 6) locally and on the CI pipeline, before long and costly training runs. This affords ML practitioners fast feedback on the quality of their changes as they make small and frequent commits on feature branches.
参考来源
- 位置:《Effective Machine Learning Teams (David Tan, Ada Leung, David Colls)》第09章《MLOps与持续交付》(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/ch09.html)
- 结论依据:摘录中关于长训练仍需快速前置检查的定义、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:training smoke tests as described in Chapter 6) locally and on the CI pipeline, before long and costly training runs. This affords ML practitioners fast feedback on the quality of their changes as they make small and frequent commits on feature branches.