← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8777 · 生成模型:GPT-6

重构护栏不能混用训练验证分数

理解等级:高级

旧脚本训练集分数与留出集分数并非同一基线。书图展示train score而特征测试使用test score,且划分和随机森林未固定随机性,不能期待0.869稳定保护;需先统一样本、度量和允许变化。

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

      assert accuracy > 0.869 We write a model metrics test in the style of a unit test. This is similar to the metrics tests that we’ve written in the earlier chapter on ML model tests. We invoke the function (which is yet to be defined) that represents the “seams” of our refactoring boundary. This function (prepare_data_and_train_model()) is the box within which we will refactor, and

    参考来源

    - 位置:《Effective Machine Learning Teams (David Tan, Ada Leung, David Colls)》第08章《重构与技术债》(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/ch08.html) - 结论依据:摘录中关于重构护栏不能混用训练验证分数的定义、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:assert accuracy > 0.869 We write a model metrics test in the style of a unit test. This is similar to the metrics tests that we’ve written in the earlier chapter on ML model tests. We invoke the function (which is yet to be defined) that represents the “seams” of our refactoring boundary. This function (prepare_data_and_train_model()) is the box within which we will refactor, and