LaTeX中实现算法的呈现主要有两种方式:
使用宏包
algorithm2e
, 这个宏包有很多可选项进行设定。使用宏包
algorithm
与algorithmic
, 好像挺多人喜欢用,周志华老师的<<机器学习>>一书中的算法描述应该就是使用的这两个宏包。
使用宏包algorithm2e
:
\usepackage[linesnumbered,boxed,ruled,commentsnumbered]{algorithm2e}%%算法包,注意设置所需可选项
\IncMargin{1em} % 使得行号不向外突出 \begin{algorithm} \SetAlgoNoLine % 不要算法中的竖线 \SetKwInOut{Input}{\textbf{输入}}\SetKwInOut{Output}{\textbf{输出}} % 替换关键词 \Input{ \\ The observed user-item pair set $S$\;\\ The feature matrix of items $F$\;\\ The content features entities $A := \{A^u,A^v\}$\;\\} \Output{ \\ $\Theta \ := \{Y^u,Y^v\}$\;\\ $W := \{W^u,W^v\}$\;\\} \BlankLine initialize the model parameter $\Theta$ and $W$ with uniform $\left(-\sqrt{6}/{k},\sqrt{6}/{k}\right)$\; % 分号 \; 区分一行结束 standarized $\Theta$\; Initialize the popularity of categories $\rho$ randomly\; \Repeat {\text{convergence}} {Draw a triple $\left(m,i,j\right)$ with 算法\ref{al2}\; \For {each latent vector $\theta \in \Theta$}{ $\theta \leftarrow \theta - \eta\frac{\partial L}{\partial \theta}$ } \For {each $W^e \in W$}{ Update $W^e$ with the rule defined in Eq.\ref{equ:W}\; } } \caption{Learning paramters for BPR\label{al3}} \end{algorithm} \DecMargin{1em}
使用宏包algorithm
与algorithmic
\usepackage{algorithm, algorithmic}
\begin{algorithm} \renewcommand{\algorithmicrequire}{\textbf{Input:}} \renewcommand{\algorithmicensure}{\textbf{Output:}} \caption{Bayesian Personalized Ranking Based Latent Feature Embedding Model} \label{alg:1} \begin{algorithmic}[1] \REQUIRE latent dimension $K$, $G$, target predicate $p$ \ENSURE $U^{p}$, $V^{p}$, $b^{p}$ \STATE Given target predicate $p$ and entire knowledge graph $G$, construct its bipartite subgraph, $G_{p}$ \STATE $m$ = number of subject entities in $G_{p}$ \STATE $n$ = number of object entities in $G_{p}$ \STATE Generate a set of training samples $D_{p} = \{(s_p, o^{+}_{p}, o^{-}_{p})\}$ using uniform sampling technique \STATE Initialize $U^{p}$ as size $m \times K$ matrix with $0$ mean and standard deviation $0.1$ \STATE Initialize $V^{p}$ as size $n \times K$ matrix with $0$ mean and stardard deviation $0.1$ \STATE Initialize $b^{p}$ as size $n \times 1$ column vector with $0$ mean and stardard deviation $0.1$ \FORALL{$(s_p, o^{+}_{p}, o^{-}_{p}) \in D_{p}$} \STATE Update $U_{s}^{p}$ based on Equation~\ref{eq:sgd1} \STATE Update $V_{o^{+}}^{p}$ based on Equation~\ref{eq:sgd2} \STATE Update $V_{o^{-}}^{p}$ based on Equation~\ref{eq:sgd3} \STATE Update $b_{o^{+}}^{p}$ based on Equation~\ref{eq:sgd4} \STATE Update $b_{o^{-}}^{p}$ based on Equation~\ref{eq:sgd5} \ENDFOR \STATE \textbf{return} $U^{p}$, $V^{p}$, $b^{p}$ \end{algorithmic} \end{algorithm}此文章转载自:https://blog.csdn.net/simple_the_best/article/details/52710794