mirror of
https://github.com/MartinThoma/LaTeX-examples.git
synced 2025-04-26 06:48:04 +02:00
50 lines
2.3 KiB
TeX
50 lines
2.3 KiB
TeX
|
\documentclass{article}
|
||
|
|
\usepackage[pdftex,active,tightpage]{preview}
|
||
|
|
\setlength\PreviewBorder{2mm}
|
||
|
|
|
||
|
|
\usepackage[utf8]{inputenc} % this is needed for umlauts
|
||
|
|
\usepackage[ngerman]{babel} % this is needed for umlauts
|
||
|
|
\usepackage[T1]{fontenc} % this is needed for correct output of umlauts in pdf
|
||
|
|
\usepackage{amssymb,amsmath,amsfonts} % nice math rendering
|
||
|
|
\usepackage{braket} % needed for \Set
|
||
|
|
\usepackage{caption}
|
||
|
|
\usepackage{algorithm}
|
||
|
|
\usepackage[noend]{algpseudocode}
|
||
|
|
|
||
|
|
\DeclareCaptionFormat{myformat}{#3}
|
||
|
|
\captionsetup[algorithm]{format=myformat}
|
||
|
|
|
||
|
|
\begin{document}
|
||
|
|
\begin{preview}
|
||
|
|
\begin{algorithm}[H]
|
||
|
|
\begin{algorithmic}
|
||
|
|
\Require
|
||
|
|
\Statex Sates $\mathcal{X} = \{1, \dots, n_x\}$
|
||
|
|
\Statex Actions $\mathcal{A} = \{1, \dots, n_a\},\qquad A: \mathcal{X} \Rightarrow \mathcal{A}$
|
||
|
|
\Statex Reward function $R: \mathcal{X} \times \mathcal{A} \rightarrow \mathbb{R}$
|
||
|
|
\Statex Black-box (probabilistic) transition function $T: \mathcal{X} \times \mathcal{A} \rightarrow \mathcal{X}$
|
||
|
|
\Statex Learning rate $\alpha \in [0, 1]$, typically $\alpha = 0.1$
|
||
|
|
\Statex Discounting factor $\gamma \in [0, 1]$
|
||
|
|
\Statex $\lambda \in [0, 1]$: Trade-off between TD and MC
|
||
|
|
\Procedure{SARSA}{$\mathcal{X}$, $A$, $R$, $T$, $\alpha$, $\gamma$, $\lambda$}
|
||
|
|
\State Initialize $Q: \mathcal{X} \times \mathcal{A} \rightarrow \mathbb{R}$ arbitrarily
|
||
|
|
\While{$Q$ is not converged}
|
||
|
|
\State Select $(s, a) \in \mathcal{X} \times \mathcal{A}$ arbitrarily
|
||
|
|
\While{$s$ is not terminal}
|
||
|
|
\State $r \gets R(s, a)$ \Comment{Receive the reward}
|
||
|
|
\State $s' \gets T(s, a)$ \Comment{Receive the new state}
|
||
|
|
\State Calculate $\pi$ based on $Q$ (e.g. epsilon-greedy)
|
||
|
|
\State $a' \gets \pi(s')$
|
||
|
|
\State $Q(s, a) \gets (1 - \alpha ) \cdot Q(s, a) + \alpha \cdot (r + \gamma Q(s', a'))$
|
||
|
|
\State $s \gets s'$
|
||
|
|
\State $a \gets a'$
|
||
|
|
\EndWhile
|
||
|
|
\EndWhile
|
||
|
|
\Return $Q$
|
||
|
|
\EndProcedure
|
||
|
|
\end{algorithmic}
|
||
|
|
\caption{SARSA: Learn function $Q: \mathcal{X} \times \mathcal{A} \rightarrow \mathbb{R}$}
|
||
|
|
\label{alg:sarsa}
|
||
|
|
\end{algorithm}
|
||
|
|
\end{preview}
|
||
|
|
\end{document}
|