← 返回刚才阅读的位置

知识卡片

结构图卡 · 8759.e · 生成模型:GPT-6

测试证据应分层互相补足

理解等级:中级

结构图:

flowchart TD
 C[代码 数据或模型变更] --> U[确定逻辑与数据契约检查]
 U --> S[小样本训练连通性]
 S --> A[本地接口与错误行为]
 A --> M[独立模型质量与场景评价]
 M --> D[真实环境部署验证]
 D --> O[生产监控与新失效反馈]
 O --> U
 O --> M

各层有不同被测责任,失败要明确阻止相应动作;全绿只表示既有检查通过,不提供覆盖以外的保证。

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

      each layer of tests prevents a specific type of issue but no single layer of tests (e.g., unit tests, training smoke tests, or metrics tests) can test for all kinds of undesired outcomes. The layers complement each other to reduce blind spots and reduce the risk of bugs, errors, and other adverse outcomes flowing into production (see Figure 5-4).

    参考来源

    - 位置:《Effective Machine Learning Teams (David Tan, Ada Leung, David Colls)》第05章《软件自动化测试》(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/ch05.html) - 结论依据:摘录中关于测试证据应分层互相补足的定义、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:each layer of tests prevents a specific type of issue but no single layer of tests (e.g., unit tests, training smoke tests, or metrics tests) can test for all kinds of undesired outcomes. The layers complement each other to reduce blind spots and reduce the risk of bugs, errors, and other adverse outcomes flowing into production (see Figure 5-4).