知识卡片
策略替换行动生成机制
策略π(a|h)依据可用历史h产生行动,部署时替换原行动机制。固定行动是特殊情况;确定性π(h)也是退化随机干预。策略依赖哪些信息,就应保留这些信息到行动的影响,而不是一律删除所有入边。
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
12.4 Policies as stochastic interventions In automated sequential decision-making, the term “policy” is preferred to “decision rule.” I’ll introduce a special notation for a policy: π(.). It will be a function that takes in observed outcomes of other variables and returns an action. To consider how a policy affects the model, we’ll contrast the DAG before and after a policy is implemented. Figure 12.23 illustrates
参考来源
- 位置:《Causal AI》第12章《因果决策与强化学习》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-12.xhtml)
- 结论依据:摘录中关于策略替换行动生成机制的定义、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:12.4 Policies as stochastic interventions In automated sequential decision-making, the term “policy” is preferred to “decision rule.” I’ll introduce a special notation for a policy: π(.). It will be a function that takes in observed outcomes of other variables and returns an action. To consider how a policy affects the model, we’ll contrast the DAG before and after a policy is implemented. Figure 12.23 illustrates