← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9247.b · 生成模型:GPT-6

L1与L2惩罚产生不同稀疏性

理解等级:中级

Lasso加λΣ|βⱼ|,可能使部分系数恰为零;Ridge加λΣβⱼ²,通常收缩而不产生同样的硬选择。尺度与截距是否惩罚要明确,λ=0对应未惩罚目标但实现和解的唯一性仍有条件。

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

      Lasso regression uses L1 regularization that adds a penalty equal to the absolute value of the magnitude of coefficients. This type of regularization is suited for models with fewer coefficients where some coefficients can become zero and are eliminated from the model. If the penalties are larger, it may result in coefficient values closer to zero which is desirable for producing simpler models. The cost

    参考来源

    - 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第06章《监督学习模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/19_chapter06.xhtml) - 结论依据:摘录中关于L1与L2惩罚产生不同稀疏性的定义、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:Lasso regression uses L1 regularization that adds a penalty equal to the absolute value of the magnitude of coefficients. This type of regularization is suited for models with fewer coefficients where some coefficients can become zero and are eliminated from the model. If the penalties are larger, it may result in coefficient values closer to zero which is desirable for producing simpler models. The cost