← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6590.c · 生成模型:GPT-6

逆概率权重要检查尾部

理解等级:高级

ATE的基本逆概率贡献为XY/e(Z)−(1−X)Y/[1−e(Z)]。得分接近0或1使少数单位权重很大,应检查重叠、有效样本量和权重分布。截断或删除样本可能改变偏差与目标;不同估计值只是诊断线索,不能单独定位因果假设错误。

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

      Mean value: 437.79246624944926 95.0% confidence interval: (358.10472302821745, 515.2480572854872) The fact that this estimator’s result differs so dramatically from the others suggest that it is relying on statistical assumptions that don’t hold in this data. Next, we’ll move on to a popular class of backdoor estimators that implement machine learning. 11.4.3 Backdoor estimation with machine learning Recent developments in causal effect estimation focus on leveraging machine

    参考来源

    - 位置:《Causal AI》第11章《构建因果推断工作流》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-11.xhtml) - 结论依据:摘录中关于逆概率权重要检查尾部的定义、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:Mean value: 437.79246624944926 95.0% confidence interval: (358.10472302821745, 515.2480572854872) The fact that this estimator’s result differs so dramatically from the others suggest that it is relying on statistical assumptions that don’t hold in this data. Next, we’ll move on to a popular class of backdoor estimators that implement machine learning. 11.4.3 Backdoor estimation with machine learning Recent developments in causal effect estimation focus on leveraging machine