知识卡片
伪点质量会改变原始模型
将确定赋值附近加很小Normal噪声,可提供可计算密度以便近似推断,但这已经加入观测或扰动模型。尺度影响后验与优化,接近零还可能造成尖锐几何和梯度问题;不能把有限尺度的结果当作原确定SCM的精确答案。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
of the exogenous variables, but for variational inference to work, we must assign the endogenous variables a distribution using pyro.sample. We could use the Dirac delta distribution, which would assign all probability value to the output of a variable’s assignment function. But gradient-based optimization won’t work in this case. Instead, we’ll approximate inference with a “pseudo-delta” distribution—a normal distribution with a very small scale parameter.
参考来源
- 位置:《Causal AI》第09章《通用反事实推断算法》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-9.xhtml)
- 结论依据:摘录中关于伪点质量会改变原始模型的定义、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:of the exogenous variables, but for variational inference to work, we must assign the endogenous variables a distribution using pyro.sample. We could use the Dirac delta distribution, which would assign all probability value to the output of a variable’s assignment function. But gradient-based optimization won’t work in this case. Instead, we’ll approximate inference with a “pseudo-delta” distribution—a normal distribution with a very small scale parameter.