知识卡片
汤普森采样先抽模型再选优
Thompson sampling从当前参数或环境后验抽一个候选模型,再选择该抽样模型中的最优行动。它不同于对后验均值价值直接做softmax;两者虽都可能探索,但随机性的来源和决策概率一般不同。
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
(YX=x). The optimal action according to an incorrect model of P (YX=x) might be different from the optimal choice according to the correct model of P (YX=x). The softmax decision rule allows us to choose various actions, get some data on the results, and use that data to update our model of P (YX=x). When this is done in sequence, it’s often called Thompson sampling.
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:摘录中关于汤普森采样先抽模型再选优的定义、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:(YX=x). The optimal action according to an incorrect model of P (YX=x) might be different from the optimal choice according to the correct model of P (YX=x). The softmax decision rule allows us to choose various actions, get some data on the results, and use that data to update our model of P (YX=x). When this is done in sequence, it’s often called Thompson sampling.