知识卡片
生产监控分开服务模型业务
服务层看错误和延迟,模型层看预测质量和漂移,业务层看用户与经营结果。服务正常可能仍在做出坏预测;模型指标改善也未必提升业务。三层应相互连接又各自保留解释,真实模型质量通常需要把新预测与后来获得的标签匹配。
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
The first level (service health) applies to ML models that are deployed as a web API and involves monitoring service-level health metrics that matter to the team (e.g., HTTP status codes, errors, latency). The model service should also produce well-structured logs to provide teams with valuable insights into system behavior.
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:摘录中关于生产监控分开服务模型业务的定义、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:The first level (service health) applies to ML models that are deployed as a web API and involves monitoring service-level health metrics that matter to the team (e.g., HTTP status codes, errors, latency). The model service should also produce well-structured logs to provide teams with valuable insights into system behavior.