知识卡片

普通读书笔记卡 · 7005.h · 生成模型:DeepSeek-V4.1-Flash

覆盖不全与伪相关:白墙检测器

理解等级:中级

训练数据可能在多个维度上「覆盖不全」:例如气压传感器在冰点以下停止工作,没有寒冷日子的数据,模型在这一带就成了盲区。这类问题尤其隐蔽,因为留出测试集按定义只是「已有数据的随机抽样」,检测不到缺失区域,只能靠缜密思考与额外采集或合成数据在验证/训练时暴露。伪相关是其特殊形式:若所有「美观」图恰好都带白色画廊墙而「不美观」的都没有,模型在留出集上准确率极高,实质只是个「白墙探测器」,上线真实数据可能一塌糊涂,需细审数据并定向探测。

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

    • Basic Introduction to Models · 查看原书依据

      Training on this data would likely result in a model that shows very high accuracy on held-out test data, but that is essentially just a white-wall detector. Deploying this model on real data could have terrible results, even while showing excellent performance on held-out test data.

    参考来源

    - 位置:《Reliable Machine Learning》第03章《Basic Introduction to Models》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch03.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:Training on this data would likely result in a model that shows very high accuracy on held-out test data, but that is essentially just a white-wall detector. Deploying this model on real data could have terrible results, even while showing excellent performance on held-out test data.