强化学习读书笔记 - 06~07 - 时序差分学习(Temporal-Difference Learning)
强化学习读书笔记 - 06~07 - 时序差分学习(Temporal-Difference Learning)
学习笔记:
Reinforcement Learning: An Introduction, Richard S. Sutton and Andrew G. Barto c 2014, 2015, 2016
数学符号看不懂的,先看看这里:
时序差分学习简话
时序差分学习结合了动态规划和蒙特卡洛方法,是强化学习的核心思想。
时序差分这个词不好理解。改为当时差分学习比较形象一些 - 表示通过当前的差分数据来学习。
蒙特卡洛的方法是模拟(或者经历)一段情节,在情节结束后,根据情节上各个状态的价值,来估计状态价值。
时序差分学习是模拟(或者经历)一段情节,每行动一步(或者几步),根据新状态的价值,然后估计执行前的状态价值。
可以认为蒙特卡洛的方法是最大步数的时序差分学习。
本章只考虑单步的时序差分学习。多步的时序差分学习在下一章讲解。
数学表示
根据我们已经知道的知识:如果可以计算出策略价值(\(\pi\)状态价值\(v_{\pi}(s)\),或者行动价值\(q_{\pi(s, a)}\)),就可以优化策略。
在蒙特卡洛方法中,计算策略的价值,需要完成一个情节(episode),通过情节的目标价值\(G_t\)来计算状态的价值。其公式:
Formula MonteCarlo
\[
V(S_t) \gets V(S_t) + \alpha \delta_t \\
\delta_t = [G_t - V(S_t)] \\
where \\
\delta_t \text{ - Monte Carlo error} \\
\alpha \text{ - learning step size}
\]
时序差分的思想是通过下一个状态的价值计算状态的价值,形成一个迭代公式(又):
Formula TD(0)
\[
V(S_t) \gets V(S_t) + \alpha \delta_t \\
\delta_t = [R_{t+1} + \gamma\ V(S_{t+1} - V(S_t)] \\
where \\
\delta_t \text{ - TD error} \\
\alpha \text{ - learning step size} \\
\gamma \text{ - reward discount rate}
\]
注:书上提出TD error并不精确,而Monte Carlo error是精确地。需要了解,在此并不拗述。
时序差分学习方法
本章介绍的是时序差分学习的单步学习方法。多步学习方法在下一章介绍。
- 策略状态价值\(v_{\pi}\)的时序差分学习方法(单步\多步)
- 策略行动价值\(q_{\pi}\)的on-policy时序差分学习方法: Sarsa(单步\多步)
- 策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法: Q-learning(单步)
- Double Q-learning(单步)
- 策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法(带importance sampling): Sarsa(多步)
- 策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法(不带importance sampling): Tree Backup Algorithm(多步)
- 策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法: \(Q(\sigma)\)(多步)
策略状态价值\(v_{\pi}\)的时序差分学习方法
单步时序差分学习方法TD(0)
- 流程图
算法描述
Initialize \(V(s)\) arbitrarily \(\forall s \in \mathcal{S}^+\)
Repeat (for each episode):
Initialize \(\mathcal{S}\)
Repeat (for each step of episode):
\(A \gets\) action given by \(\pi\) for \(S\)
Take action \(A\), observe \(R, S'\)
\(V(S) \gets V(S) + \alpha [R + \gamma V(S') - V(S)]\)
\(S \gets S'\)
Until S is terminal
多步时序差分学习方法
- 流程图
算法描述
Input: the policy \(\pi\) to be evaluated
Initialize \(V(s)\) arbitrarily \(\forall s \in \mathcal{S}\)
Parameters: step size \(\alpha \in (0, 1]\), a positive integer \(n\)
All store and access operations (for \(S_t\) and \(R_t\)) can take their index mod \(n\)Repeat (for each episode):
Initialize and store \(S_0 \ne terminal\)
\(T \gets \infty\)
For \(t = 0,1,2,\cdots\):
If \(t < T\), then:
Take an action according to \(\pi(\dot | S_t)\)
Observe and store the next reward as \(R_{t+1}\) and the next state as \(S_{t+1}\)
If \(S_{t+1}\) is terminal, then \(T \gets t+1\)
\(\tau \gets t - n + 1 \ \) (\(\tau\) is the time whose state's estimate is being updated)
If \(\tau \ge 0\):
\(G \gets \sum_{i = \tau + 1}^{min(\tau + n, T)} \gamma^{i-\tau-1}R_i\)
if \(\tau + n \le T\) then: \(G \gets G + \gamma^{n}V(S_{\tau + n}) \qquad \qquad (G_{\tau}^{(n)})\)
\(V(S_{\tau}) \gets V(S_{\tau}) + \alpha [G - V(S_{\tau})]\)
Until \(\tau = T - 1\)
这里要理解\(V(S_0)\)是由\(V(S_0), V(S_1), \dots, V(S_n)\)计算所得;\(V(S_1)\)是由\(V(S_1), V(S_1), \dots, V(S_{n+1})\)。
策略行动价值\(q_{\pi}\)的on-policy时序差分学习方法: Sarsa
单步时序差分学习方法
- 流程图
算法描述
Initialize \(Q(s, a), \forall s \in \mathcal{S}, a \in \mathcal{A}(s)\) arbitrarily, and \(Q(termnal-state, \dot) = 0\)
Repeat (for each episode):
Initialize \(\mathcal{S}\)
Choose \(A\) from \(S\) using policy derived from \(Q\) (e.g. \(\epsilon-greedy\))
Repeat (for each step of episode):
Take action \(A\), observe \(R, S'\)
Choose \(A'\) from \(S'\) using policy derived from \(Q\) (e.g. \(\epsilon-greedy\))
\(Q(S, A) \gets Q(S, A) + \alpha [R + \gamma Q(S', A') - Q(S, A)]\)
\(S \gets S'; A \gets A';\)
Until S is terminal
多步时序差分学习方法
- 流程图
算法描述
Initialize \(Q(s, a)\) arbitrarily \(\forall s \in \mathcal{S}^, \forall a in \mathcal{A}\)
Initialize \(\pi\) to be \(\epsilon\)-greedy with respect to Q, or to a fixed given policy
Parameters: step size \(\alpha \in (0, 1]\),
small \(\epsilon > 0\)
a positive integer \(n\)
All store and access operations (for \(S_t\) and \(R_t\)) can take their index mod \(n\)Repeat (for each episode):
Initialize and store \(S_0 \ne terminal\)
Select and store an action \(A_0 ~ \pi(\dot | S_0)\)
\(T \gets \infty\)
For \(t = 0,1,2,\cdots\):
If \(t < T\), then:
Take an action \(A_t\)
Observe and store the next reward as \(R_{t+1}\) and the next state as \(S_{t+1}\)
If \(S_{t+1}\) is terminal, then:
\(T \gets t+1\)
Else:
Select and store an action \(A_{t+1} ~ \pi(\dot | S_{t+1})\)
\(\tau \gets t - n + 1 \ \) (\(\tau\) is the time whose state's estimate is being updated)
If \(\tau \ge 0\):
\(G \gets \sum_{i = \tau + 1}^{min(\tau + n, T)} \gamma^{i-\tau-1}R_i\)
if \(\tau + n \le T\) then: \(G \gets G + \gamma^{n} Q(S_{\tau + n}, A_{\tau + n}) \qquad \qquad (G_{\tau}^{(n)})\)
\(Q(S_{\tau}, A_{\tau}) \gets Q(S_{\tau}, A_{\tau}) + \alpha [G - Q(S_{\tau}, A_{\tau})]\)
If {\pi} is being learned, then ensure that \(\pi(\dot | S_{\tau})\) is \(\epsilon\)-greedy wrt Q
Until \(\tau = T - 1\)
策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法: Q-learning
Q-learning 算法(Watkins, 1989)是一个突破性的算法。这里利用了这个公式进行off-policy学习。
\[
Q(S_t, A_t) \gets Q(S_t, A_t) + \alpha [R_{t+1} + \gamma \underset{a}{max} \ Q(S_{t+1}, a) - Q(S_t, A_t)]
\]
单步时序差分学习方法
算法描述
Initialize \(Q(s, a), \forall s \in \mathcal{S}, a \in \mathcal{A}(s)\) arbitrarily, and \(Q(termnal-state, \dot) = 0\)
Repeat (for each episode):
Initialize \(\mathcal{S}\)
Choose \(A\) from \(S\) using policy derived from \(Q\) (e.g. \(\epsilon-greedy\))
Repeat (for each step of episode):
Take action \(A\), observe \(R, S'\)
\(Q(S, A) \gets Q(S, A) + \alpha [R + \gamma \underset{a}{max} \ Q(S‘, a) - Q(S, A)]\)
\(S \gets S';\)
Until S is terminalQ-learning使用了max,会引起一个最大化偏差(Maximization Bias)问题。
具体说明,请看书上的Example 6.7。**
使用Double Q-learning可以消除这个问题。
Double Q-learning
单步时序差分学习方法
Initialize $Q_1(s, a) and \(Q_2(s, a), \forall s \in \mathcal{S}, a \in \mathcal{A}(s)\) arbitrarily
Initialize $Q_1(termnal-state, \dot) = \(Q_2(termnal-state, \dot) = 0\)
Repeat (for each episode):
Initialize \(\mathcal{S}\)
Repeat (for each step of episode):
Choose \(A\) from \(S\) using policy derived from \(Q_1\) and \(Q_2\) (e.g. \(\epsilon-greedy\))
Take action \(A\), observe \(R, S'\)
With 0.5 probability:
\(Q_1(S, A) \gets Q_1(S, A) + \alpha [R + \gamma Q_2(S', \underset{a}{argmax} \ Q_1(S', a)) - Q_1(S, A)]\)
Else:
\(Q_2(S, A) \gets Q_2(S, A) + \alpha [R + \gamma Q_1(S', \underset{a}{argmax} \ Q_2(S', a)) - Q_2(S, A)]\)
\(S \gets S';\)
Until S is terminal
策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法(by importance sampling): Sarsa
考虑到重要样本,把\(\rho\)带入到Sarsa算法中,形成一个off-policy的方法。
\(\rho\) - 重要样本比率(importance sampling ratio)
\[
\rho \gets \Pi_{i = \tau + 1}^{min(\tau + n - 1, T -1 )} \frac{\pi(A_t|S_t)}{\mu(A_t|S_t)} \qquad \qquad (\rho_{\tau+n}^{(\tau+1)})
\]
多步时序差分学习方法
算法描述
Input: behavior policy \mu such that \(\mu(a|s) > 0,\forall s \in \mathcal{S}, a \in \mathcal{A}\)
Initialize \(Q(s,a)\) arbitrarily \(\forall s \in \mathcal{S}^, \forall a in \mathcal{A}\)
Initialize \(\pi\) to be \(\epsilon\)-greedy with respect to Q, or to a fixed given policy
Parameters: step size \(\alpha \in (0, 1]\),
small \(\epsilon > 0\)
a positive integer \(n\)
All store and access operations (for \(S_t\) and \(R_t\)) can take their index mod \(n\)Repeat (for each episode):
Initialize and store \(S_0 \ne terminal\)
Select and store an action \(A_0 ~ \mu(\dot | S_0)\)
\(T \gets \infty\)
For \(t = 0,1,2,\cdots\):
If \(t < T\), then:
Take an action \(A_t\)
Observe and store the next reward as \(R_{t+1}\) and the next state as \(S_{t+1}\)
If \(S_{t+1}\) is terminal, then:
\(T \gets t+1\)
Else:
Select and store an action \(A_{t+1} ~ \pi(\dot | S_{t+1})\)
\(\tau \gets t - n + 1 \ \) (\(\tau\) is the time whose state's estimate is being updated)
If \(\tau \ge 0\):
\(\rho \gets \Pi_{i = \tau + 1}^{min(\tau + n - 1, T -1 )} \frac{\pi(A_t|S_t)}{\mu(A_t|S_t)} \qquad \qquad (\rho_{\tau+n}^{(\tau+1)})\)
\(G \gets \sum_{i = \tau + 1}^{min(\tau + n, T)} \gamma^{i-\tau-1}R_i\)
if \(\tau + n \le T\) then: \(G \gets G + \gamma^{n} Q(S_{\tau + n}, A_{\tau + n}) \qquad \qquad (G_{\tau}^{(n)})\)
\(Q(S_{\tau}, A_{\tau}) \gets Q(S_{\tau}, A_{\tau}) + \alpha \rho [G - Q(S_{\tau}, A_{\tau})]\)
If {\pi} is being learned, then ensure that \(\pi(\dot | S_{\tau})\) is \(\epsilon\)-greedy wrt Q
Until \(\tau = T - 1\)
Expected Sarsa
- 流程图
- 算法描述
略。
策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法(不带importance sampling): Tree Backup Algorithm
Tree Backup Algorithm的思想是每步都求行动价值的期望值。
求行动价值的期望值意味着对所有可能的行动\(a\)都评估一次。
多步时序差分学习方法
- 流程图
算法描述
Initialize \(Q(s,a)\) arbitrarily \(\forall s \in \mathcal{S}^, \forall a in \mathcal{A}\)
Initialize \(\pi\) to be \(\epsilon\)-greedy with respect to Q, or to a fixed given policy
Parameters: step size \(\alpha \in (0, 1]\),
small \(\epsilon > 0\)
a positive integer \(n\)
All store and access operations (for \(S_t\) and \(R_t\)) can take their index mod \(n\)Repeat (for each episode):
Initialize and store \(S_0 \ne terminal\)
Select and store an action \(A_0 ~ \pi(\dot | S_0)\)
\(Q_0 \gets Q(S_0, A_0)\)
\(T \gets \infty\)
For \(t = 0,1,2,\cdots\):
If \(t < T\), then:
Take an action \(A_t\)
Observe and store the next reward as \(R_{t+1}\) and the next state as \(S_{t+1}\)
If \(S_{t+1}\) is terminal, then:
\(T \gets t+1\)
\(\delta_t \gets R - Q_t\)
Else:
\(\delta_t \gets R + \gamma \sum_a \pi(a|S_{t+1})Q(S_{t+1},a) - Q_t\)
Select arbitrarily and store an action as \(A_{t+1}\)
\(Q_{t+1} \gets Q(S_{t+1},A_{t+1})\)
\(\pi_{t+1} \gets \pi(S_{t+1},A_{t+1})\)
\(\tau \gets t - n + 1 \ \) (\(\tau\) is the time whose state's estimate is being updated)
If \(\tau \ge 0\):
\(E \gets 1\)
\(G \gets Q_{\tau}\)
For \(k=\tau, \dots, min(\tau + n - 1, T - 1):\)
\(G \gets\ G + E \delta_k\)
\(E \gets\ \gamma E \pi_{k+1}\)
\(Q(S_{\tau}, A_{\tau}) \gets Q(S_{\tau}, A_{\tau}) + \alpha [G - Q(S_{\tau}, A_{\tau})]\)
If {\pi} is being learned, then ensure that \(\pi(a | S_{\tau})\) is \(\epsilon\)-greedy wrt \(Q(S_{\tau},\dot)\)
Until \(\tau = T - 1\)
策略行动价值\(q_{\pi}\)的off-policy时序差分学习方法: \(Q(\sigma)\)
\(Q(\sigma)\)结合了Sarsa(importance sampling), Expected Sarsa, Tree Backup算法,并考虑了重要样本。
当\(\sigma = 1\)时,使用了重要样本的Sarsa算法。
当\(\sigma = 0\)时,使用了Tree Backup的行动期望值算法。
多步时序差分学习方法
- 流程图
算法描述
Input: behavior policy \mu such that \(\mu(a|s) > 0,\forall s \in \mathcal{S}, a \in \mathcal{A}\)
Initialize \(Q(s,a)\) arbitrarily \(\forall s \in \mathcal{S}^, \forall a in \mathcal{A}\)
Initialize \(\pi\) to be \(\epsilon\)-greedy with respect to Q, or to a fixed given policy
Parameters: step size \(\alpha \in (0, 1]\),
small \(\epsilon > 0\)
a positive integer \(n\)
All store and access operations (for \(S_t\) and \(R_t\)) can take their index mod \(n\)Repeat (for each episode):
Initialize and store \(S_0 \ne terminal\)
Select and store an action \(A_0 ~ \mu(\dot | S_0)\)
\(Q_0 \gets Q(S_0, A_0)\)
\(T \gets \infty\)
For \(t = 0,1,2,\cdots\):
If \(t < T\), then:
Take an action \(A_t\)
Observe and store the next reward as \(R_{t+1}\) and the next state as \(S_{t+1}\)
If \(S_{t+1}\) is terminal, then:
\(T \gets t+1\)
\(\delta_t \gets R - Q_t\)
Else:
Select and store an action as \(A_{t+1} ~ \mu(\dot|S_{t+1})\)
Select and store \(\sigma_{t+1})\)
\(Q_{t+1} \gets Q(S_{t+1},A_{t+1})\)
\(\delta_t \gets R + \gamma \sigma_{t+1} Q_{t+1} + \gamma (1 - \sigma_{t+1})\sum_a \pi(a|S_{t+1})Q(S_{t+1},a) - Q_t\)
\(\pi_{t+1} \gets \pi(S_{t+1},A_{t+1})\)
\(\rho_{t+1} \gets \frac{\pi(A_{t+1}|S_{t+1})}{\mu(A_{t+1}|S_{t+1})}\)
\(\tau \gets t - n + 1 \ \) (\(\tau\) is the time whose state's estimate is being updated)
If \(\tau \ge 0\):
\(\rho \gets 1\)
\(E \gets 1\)
\(G \gets Q_{\tau}\)
For \(k=\tau, \dots, min(\tau + n - 1, T - 1):\)
\(G \gets\ G + E \delta_k\)
\(E \gets\ \gamma E [(1 - \sigma_{k+1})\pi_{k+1} + \sigma_{k+1}]\)
\(\rho \gets\ \rho(1 - \sigma_{k} + \sigma_{k}\tau_{k})\)
$Q(S_{\tau}, A_{\tau}) \gets Q(S_{\tau}, A_{\tau}) + \alpha \(\rho [G - Q(S_{\tau}, A_{\tau})]\)
If {\pi} is being learned, then ensure that \(\pi(a | S_{\tau})\) is \(\epsilon\)-greedy wrt \(Q(S_{\tau},\dot)\)
Until \(\tau = T - 1\)
总结
时序差分学习方法的限制:学习步数内,可获得奖赏信息。
比如,国际象棋的每一步,是否可以计算出一个奖赏信息?如果使用蒙特卡洛方法,模拟到游戏结束,肯定是可以获得一个奖赏结果的。
参照
- Reinforcement Learning: An Introduction, Richard S. Sutton and Andrew G. Barto c 2014, 2015, 2016
- 强化学习读书笔记 - 00 - 术语和数学符号
- 强化学习读书笔记 - 01 - 强化学习的问题
- 强化学习读书笔记 - 02 - 多臂老O虎O机问题
- 强化学习读书笔记 - 03 - 有限马尔科夫决策过程
- 强化学习读书笔记 - 04 - 动态规划
- 强化学习读书笔记 - 05 - 蒙特卡洛方法(Monte Carlo Methods)
强化学习读书笔记 - 06~07 - 时序差分学习(Temporal-Difference Learning)的更多相关文章
- Python深度学习读书笔记-1.什么是深度学习
人工智能 什么是人工智能.机器学习与深度学习(见图1-1)?这三者之间有什么关系?
- 强化学习读书笔记 - 13 - 策略梯度方法(Policy Gradient Methods)
强化学习读书笔记 - 13 - 策略梯度方法(Policy Gradient Methods) 学习笔记: Reinforcement Learning: An Introduction, Richa ...
- 强化学习读书笔记 - 12 - 资格痕迹(Eligibility Traces)
强化学习读书笔记 - 12 - 资格痕迹(Eligibility Traces) 学习笔记: Reinforcement Learning: An Introduction, Richard S. S ...
- 强化学习读书笔记 - 11 - off-policy的近似方法
强化学习读书笔记 - 11 - off-policy的近似方法 学习笔记: Reinforcement Learning: An Introduction, Richard S. Sutton and ...
- 强化学习读书笔记 - 10 - on-policy控制的近似方法
强化学习读书笔记 - 10 - on-policy控制的近似方法 学习笔记: Reinforcement Learning: An Introduction, Richard S. Sutton an ...
- 强化学习读书笔记 - 09 - on-policy预测的近似方法
强化学习读书笔记 - 09 - on-policy预测的近似方法 参照 Reinforcement Learning: An Introduction, Richard S. Sutton and A ...
- 强化学习读书笔记 - 02 - 多臂老O虎O机问题
# 强化学习读书笔记 - 02 - 多臂老O虎O机问题 学习笔记: [Reinforcement Learning: An Introduction, Richard S. Sutton and An ...
- 强化学习读书笔记 - 05 - 蒙特卡洛方法(Monte Carlo Methods)
强化学习读书笔记 - 05 - 蒙特卡洛方法(Monte Carlo Methods) 学习笔记: Reinforcement Learning: An Introduction, Richard S ...
- 深度学习读书笔记之RBM(限制波尔兹曼机)
深度学习读书笔记之RBM 声明: 1)看到其他博客如@zouxy09都有个声明,老衲也抄袭一下这个东西 2)该博文是整理自网上很大牛和机器学习专家所无私奉献的资料的.具体引用的资料请看参考文献.具体的 ...
随机推荐
- 上传预览 easyui部分控件获取focuse 表单验证
js: $(document).ready(function () { //$('#creater').combobox({ // url: '/VMS.UI/BindData/ScheamData? ...
- hyperLink的定制
在iReport中增加hyperLink,点击之后没有反应:需要重新写一遍net.sf.jasperreports.view.JRViewer; 修改其中的gotoHyperlink方法: case ...
- Sql Server 查询多行并一行
干货 CREATE TABLE #benefit_code21 (id INT, number nvarchar(MAX), pname ), collegeID INT, applicationda ...
- enum 用法
public enum WeekDay { SUN(, "Sunday", "SUN"), MON(, "Monday", "MO ...
- Nancy简单实战之NancyMusicStore(六):写在最后
前言 由于公司搬家后,住的地方离上班的地方远了N倍,以前是走路十多分钟就可以到公司的,上班时间也从9:00提早到8:30 现在每天上班都是先坐公交,然后再坐地铁,在这段路上比较浪费时间而且每天都是要6 ...
- HDU4496(并查集)
D-City Time Limit: 2000/1000 MS (Java/Others) Memory Limit: 65535/65535 K (Java/Others)Total Subm ...
- 【python基础】 Tkinter小构件之canvas 画布
[python之tkinter画布] 要画布就要使用tkinter的小构件,canvas(结构化的图形,用于绘制图形,创建图形编辑器以及实现自定制的小构件类) 我们先使用create_rectangl ...
- leetcode刷题总结
题外话 今年大三,现正值寒假时间,开学就开始大三下学期的生活了. 在大三临近结束的时间,也就是复习考试的时间里,我每天都会用早上的时间来刷codewars.刚开始玩的时候,一到8kyu的题目如果稍微难 ...
- Docker 1.13 管理命令
1.12 CLI 的问题 Docker1.12 命令行接口(CLI)有40多个顶级命令,这些命令存在以下问题: 没有归类组织,这让docker 新手很难学习: 有些命令没有提供足够的上下文环境,以至于 ...
- 不是技术牛人,如何拿到国内IT巨头的Offer (转载)
原文地址:http://blog.csdn.net/lsldd/article/details/13506263 不久前,byvoid面阿里星计划的面试结果截图泄漏,引起无数IT屌丝的羡慕敬仰.看看这 ...