Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix some typo and Change ^M to \n in some file #24

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Fix some typo and Change ^M to \n in some file #24

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
01f2383
fix some typo and change ^M to \n and fix hspace bug in page.tex
ycsos Aug 16, 2021
90184c3
update tensorflow-internals.pdf, you can make from this latex source …
ycsos Aug 16, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 9 additions & 9 deletions contents/architecture.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,7 +131,7 @@ \subsection{图构造}
\begin{figure}[!htbp]
\centering
\includegraphics[width=0.9\textwidth]{figures/tf-graph-construction.png}
\caption{图构造}}
\caption{图构造}
\label{fig:tf-graph-construction}
\end{figure}

Expand All @@ -144,7 +144,7 @@ \subsection{图执行}
\begin{figure}[!htbp]
\centering
\includegraphics[width=0.9\textwidth]{figures/tf-graph-execution.png}
\caption{图执行}}
\caption{图执行}
\label{fig:tf-graph-execution}
\end{figure}

Expand All @@ -155,7 +155,7 @@ \subsubsection{图分裂}
\begin{figure}[!htbp]
\centering
\includegraphics[width=1.0\textwidth]{figures/tf-graph-split-by-task.png}
\caption{图分裂：按任务划分}}
\caption{图分裂：按任务划分}
\label{fig:tf-graph-split-by-task}
\end{figure}

Expand All @@ -170,7 +170,7 @@ \subsubsection{子图注册}
\begin{figure}[!htbp]
\centering
\includegraphics[width=1.0\textwidth]{figures/tf-register-graph.png}
\caption{子图注册：插入Send和Recv节点}}
\caption{子图注册：插入Send和Recv节点}
\label{fig:tf-register-graph}
\end{figure}

Expand All @@ -181,7 +181,7 @@ \subsubsection{子图运算}
\begin{figure}[!htbp]
\centering
\includegraphics[width=1.0\textwidth]{figures/tf-run-graph.png}
\caption{子图执行}}
\caption{子图执行}
\label{fig:tf-run-graph}
\end{figure}

Expand All @@ -202,7 +202,7 @@ \subsection{创建会话}
\begin{figure}[!h]
\centering
\includegraphics[width=0.7\textwidth]{figures/tf-create-session-overview.png}
\caption{创建会话}}
\caption{创建会话}
\label{fig:tf-create-session-overview}
\end{figure}

Expand All @@ -213,7 +213,7 @@ \subsection{迭代运行}
\begin{figure}[!h]
\centering
\includegraphics[width=1.0\textwidth]{figures/tf-run-step-overview.png}
\caption{迭代执行}}
\caption{迭代执行}
\label{fig:tf-run-step-overview}
\end{figure}

Expand Down Expand Up @@ -251,7 +251,7 @@ \subsubsection{交换数据}
\begin{figure}[!h]
\centering
\includegraphics[width=0.7\textwidth]{figures/tf-recv-tensor-overview.png}
\caption{Worker之间的数据交换}}
\caption{Worker之间的数据交换}
\label{fig:tf-recv-tensor-overview}
\end{figure}

Expand All @@ -262,7 +262,7 @@ \subsection{关闭会话}
\begin{figure}[!h]
\centering
\includegraphics[width=0.7\textwidth]{figures/tf-close-session-overview.png}
\caption{关闭会话}}
\caption{关闭会话}
\label{fig:tf-close-session-overview}
\end{figure}

Expand Down
2 changes: 1 addition & 1 deletion contents/code-reading.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,7 +228,7 @@ \section{实例化}

\item \ascii{Stage 1}的划分
\begin{enum}
\eitem{\ascii{A}没有父亲}RDD}，\ascii{Stage(1)}划分结束。特殊地\ascii{Stage(1)}仅包含\ascii{RDD A}。}
\eitem{\ascii{A}没有父亲\ascii{RDD}，\ascii{Stage(1)}划分结束。特殊地\ascii{Stage(1)}仅包含\ascii{RDD A}。}
\end{enum}
\item \ascii{Stage 2}的划分
\begin{enum}
Expand Down
24 changes: 11 additions & 13 deletions contents/ice-breaker.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,8 +51,6 @@ \subsection{样本数据集}

\subsection{图示说明}

\begin{content}

为了更好地可视化整个训练过程，使用\ascii{matplotlib}工具包绘制了\ascii{5}种类型的画板。如\refig{mnist-training-digits}所示，表示一个\ascii{mini-batch}的训练样本数据集。其中，\code{batch\_size = 100}，白色背景表示数字被正确识别；而红色背景表示数字被误分类，手写数字的左侧标识了正确的标签值，而右侧标识了错误的预测值。

\ascii{MNIST}拥有\ascii{50000}个训练样本，如果\code{batch\_size}为\ascii{100}，则需要迭代\ascii{500}次才能完整地遍历一次训练样本数据集，常称为一个\ascii{epoch}周期。
Expand Down Expand Up @@ -414,21 +412,21 @@ \subsubsection{符号定义}
为了形式化地描述多层感知器模型，此处定义了一些常用符号。

\begin{itemize}
\item \alert{$ {n_{\ell}} $}: 网络层数，其中第$0$层为输入层，第$n_{\ell}$层为输出层
\item \alert{$ {s_{\ell}} $}: 第$\ell$层的节点数，$ \ell = 0, 1, ..., n_{\ell} $
\item \alert{$ w_{ji}^{(\ell)} $}: 第$(\ell-1)$层节点$i$与第$\ell$层节点$j$之间的权重，$ \ell = 1, ..., n_{\ell} $
\item \alert{$ b_i^{(\ell)} $}: 第$\ell$层节点$i$的偏置项，$ \ell = 1, ..., n_{\ell} $
\item \alert{$ a_i^{(\ell)} $}: 第$\ell$层节点$i$的输出，$ \ell = 1, ..., n_{\ell}, x = a^{(0)}, y = a^{(n_{\ell})} $
\item \alert{$ z_i^{(\ell)} $}: 第$\ell$层节点$i$的权重和，$ \ell = 1, ..., n_{\ell} $
\item \alert{$ \delta _i^{(\ell)} $}: 第$\ell$层节点$i$的误差项，$ \ell = 1, ..., n_{\ell} $
\item \alert{$ S = \{ ({x^{(t)}},{y^{(t)}});t = 1,2,...,m\} $}: 样本空间
\item $ {n_{\ell}} $: 网络层数，其中第$0$层为输入层，第$n_{\ell}$层为输出层
\item $ {s_{\ell}} $: 第$\ell$层的节点数，$ \ell = 0, 1, ..., n_{\ell} $
\item $ w_{ji}^{(\ell)} $: 第$(\ell-1)$层节点$i$与第$\ell$层节点$j$之间的权重，$ \ell = 1, ..., n_{\ell} $
\item $ b_i^{(\ell)} $: 第$\ell$层节点$i$的偏置项，$ \ell = 1, ..., n_{\ell} $
\item $ a_i^{(\ell)} $: 第$\ell$层节点$i$的输出，$ \ell = 1, ..., n_{\ell}, x = a^{(0)}, y = a^{(n_{\ell})} $
\item $ z_i^{(\ell)} $: 第$\ell$层节点$i$的权重和，$ \ell = 1, ..., n_{\ell} $
\item $ \delta _i^{(\ell)} $: 第$\ell$层节点$i$的误差项，$ \ell = 1, ..., n_{\ell} $
\item $ S = \{ ({x^{(t)}},{y^{(t)}});t = 1,2,...,m\} $: 样本空间
\end{itemize}

\subsubsection{前向传播}

$z^{(\ell )}$表示$\ell$层的线性加权和，它由第$\ell - 1$层的输出$a^{(\ell - 1)}$与第$\ell$层的权重矩阵$w^{(\ell )}$相乘，再加上第$\ell$层的偏置向量所得。

推而广之，第$\ell$层的输出，由激活函数$f({z^{(\ell )}})$所得。其中，$a^{(0)}} = x, y = {a^{({n_\ell })}}$。
推而广之，第$\ell$层的输出，由激活函数$f({z^{(\ell )}})$所得。其中，$a^{(0)} = x, y = {a^{({n_\ell })}}$。

\[\begin{gathered}
{z^{(\ell )}} = {w^{(\ell )}}{a^{(\ell - 1)}} + {b^{(\ell )}} \hfill \\
Expand Down Expand Up @@ -804,7 +802,7 @@ \subsubsection{前向传播}

$Z^{(\ell )}$表示$\ell$层的线性加权和，它由第$\ell - 1$层的输出$A^{(\ell - 1)}$与第$\ell$层的权重矩阵$W^{(\ell )}$卷积，再加上第$\ell$层的偏置向量所得。

推而广之，第$\ell$层的输出，由激活函数$f({Z^{(\ell )}})$所得。其中，$A^{(0)}} = x, y = {A^{({n_\ell })}}$。
推而广之，第$\ell$层的输出，由激活函数$f({Z^{(\ell )}})$所得。其中，$A^{(0)} = x, y = {A^{({n_\ell })}}$。

\[\begin{gathered}
{Z^{(\ell )}} = {A^{(\ell - 1)}} * {W^{(\ell )}} + {b^{(\ell )}} \hfill \\
Expand Down Expand Up @@ -969,4 +967,4 @@ \subsection{增强卷积网络}
\label{fig:mnist-conv2d-3-result}
\end{figure}

\end{content}
\end{content}
2 changes: 2 additions & 0 deletions contents/input-pipeline.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -387,3 +387,5 @@ \subsection{Pipeline节拍}
同样的道理，下游\ascii{OP}从队列(文件名队列)中出队元素，一旦该队列元素为空，则自动抛出\code{OutOfRangeError}异常。该阶段对应的\code{QueueRunner}将感知该异常的发生，然后捕获异常并关闭下游的队列(样本队列)，退出线程的执行。

在\ascii{Pipeline}的最后阶段，\code{train\_op}从样本队列中出队批次训练样本时，队列为空，并且队列被关闭了，则抛出\code{OutOfRangeError}异常，最终停止整个训练任务。

\end{content}
9 changes: 9 additions & 0 deletions contents/local.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,8 +47,12 @@ \subsection{并发执行}

也就是说，分裂和执行体现了\tf{}并发执行的设计理念。

\end{content}

\section{会话控制}

\begin{content}

在本地模式下，其运行时由\code{DirectSession}控制。一般地，\code{DirectSession}执行计算图时，各组件之间都是函数调用关系。但是，\code{DirectSession}也存在清晰的生命周期管理机制，如\refig{local-direct-session-lifecycle}所示。

\begin{figure}[H]
Expand Down Expand Up @@ -418,10 +422,13 @@ \subsection{关闭会话}
\end{c++}
\end{leftbar}

\end{content}

\section{剪枝}
\label{sec:graph-operation-prune}

\begin{content}

\code{DirectSession::Run}执行时，首先完成\code{ClientGraph}的构造。事实上，\code{ClientGraph}的构造过程，主要完成\code{FullGraph}的剪枝算法，并生成\code{ClientGraph}。

\subsection{构建ClientGraph}
Expand Down Expand Up @@ -851,6 +858,8 @@ \subsubsection{反向剪枝}
\end{c++}
\end{leftbar}

\end{content}

\section{分裂}
\label{sec:graph-operation-split}

Expand Down
10 changes: 5 additions & 5 deletions contents/prog-env.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -534,7 +534,7 @@ \subsubsection{创建Eclipse工程}

创建一个\ascii{Eclipse C++}工程，如\refig{setup-eclipse}所示。确定唯一的项目名称，手动地指定\ascii{TensorFlow}源代码的根目录，并选择\ascii{Makefile}的空工程。然后，按照\ascii{Properties > C/C++ General > Paths and Symbols > Includes}配置头文件的搜索目录。

\begin{table}[!htbl]
\begin{table}[!htbp]
\resizebox{0.95\textwidth}{!} {
\begin{tabular*}{1.2\textwidth}{@{}ll@{}}
\toprule
Expand All @@ -549,18 +549,18 @@ \subsubsection{创建Eclipse工程}
\label{tbl:tf-includes}
\end{table}

\begin{figure}[!htbl]
\begin{figure}[!htbp]
\centering
\includegraphics[width=0.75\textwidth]{figures/setup-eclipse.png}
\caption{创建Eclipse C++工程}}
\caption{创建Eclipse C++工程}
\label{fig:setup-eclipse}
\end{figure}

\subsubsection{配置Eigen}

不幸的是，\ascii{Eigen}对外公开的头文件缺少\code{.h}的后缀名，\ascii{CDT}无法解析相关的符号。请参阅\code{\href{http://eigen.tuxfamily.org/index.php?title=IDEs}{http://eigen.tuxfamily.org/index.php?title=IDEs}}相关说明，按照\ascii{Preferences > C/C++ > Coding Style > Organize Includes > Header Substitution}导入\code{eigen-header-substitution.xml}文件，如\refig{eclipse-eigen3}所示。

\begin{figure}[!htbl]
\begin{figure}[!htbp]
\centering
\includegraphics[width=0.75\textwidth]{figures/eclipse-eigen3.png}
\caption{替换\ascii{Eigen}的头文件}
Expand All @@ -586,7 +586,7 @@ \section{技术栈}
\begin{figure}[H]
\centering
\includegraphics[width=0.7\textwidth]{figures/tf-stack.png}
\caption{TensorFlow技术栈}}
\caption{TensorFlow技术栈}
\label{fig:tf-stack}
\end{figure}

Expand Down
2 changes: 2 additions & 0 deletions contents/queue.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -289,3 +289,5 @@ \subsubsection{监听队列关闭}
\subsection{关闭队列}

当队列被关闭后，对于任何尝试\code{Enqueue}将会产生错误。但是，对于任何尝试\code{Dequeue}依然是成功的，只要队列中遗留元素；否则，\code{Dequeue}将立即失败，抛出\code{OutOfRangeError}异常，而不会阻塞等待更多元素被入队。

\end{content}
3 changes: 2 additions & 1 deletion contents/session.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,6 +139,7 @@ \subsubsection{销毁会话}
}
\end{c++}
\end{leftbar}
\end{content}

\section{默认会话}

Expand Down Expand Up @@ -366,4 +367,4 @@ \subsubsection{销毁会话}
\end{python}
\end{leftbar}

\begin{content}
\end{content}
117 changes: 116 additions & 1 deletion style/command.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1,116 @@
\newenvironment{enum}{ \begin{spacing}{1.2} % \begin{enumerate}[1.] \begin{enumerate} \setlength{\itemsep}{0pt} \setlength{\itemindent}{2em} %\setlength{\listparindent}{2em}}{% \end{enumerate} \end{spacing}}\newenvironment{nitemize}{ \begin{itemize} \setlength{\itemsep}{0pt} \setlength{\itemindent}{2em}}{% \end{itemize}}\newcommand{\suggest}[1]{\tikzstyle{mybox} = [draw=black, very thick,rectangle, rounded corners, inner sep=9pt, inner ysep=20pt]\tikzstyle{fancytitle} =[fill=white, text=black, ellipse]\noindent\begin{tikzpicture}\node [mybox] (box){%\begin{minipage}{\textwidth}\fangsong#1\end{minipage}};\node[fancytitle, right=10pt] at (box.north west) {\emph{建议}};% \node[fancytitle, rounded corners] at (box.east) {$\clubsuit$};\end{tikzpicture}}\newcommand{\notice}[1]{\tikzstyle{mybox} = [draw=black, very thick,rectangle, rounded corners, inner sep=9pt, inner ysep=20pt]\tikzstyle{fancytitle} =[fill=white, text=black]\noindent\begin{tikzpicture}\node [mybox] (box){%\begin{minipage}{\textwidth}\fangsong#1\end{minipage}};\node[fancytitle, right=10pt] at (box.north west) {\emph{注意}};%\node[fancytitle, rounded corners] at (box.east) {$\clubsuit$};\end{tikzpicture}}\newcommand{\tip}[1]{\tikzstyle{mybox} = [draw=black, very thick,rectangle, rounded corners, inner sep=9pt, inner ysep=20pt]\tikzstyle{fancytitle} =[fill=white, text=black]\noindent\begin{tikzpicture}\node [mybox] (box){%\begin{minipage}{\textwidth}\fangsong#1\end{minipage}};\node[fancytitle, right=10pt] at (box.north west) {\emph{提示}};%\node[fancytitle, rounded corners] at (box.east) {$\clubsuit$};\end{tikzpicture}}\newcommand\refch[1]{\ascii{第\ref{ch:#1}章(\nameref{ch:#1})}}\newcommand\refsec[1]{\ascii{\ref{sec:#1}节(\nameref{sec:#1})}}\newcommand\eitem[1]{\item {\itshape {#1}}}\newcommand\cpp{\ascii{C\nobreak+\nobreak+}}\newcommand\clang{\ascii{C}}\newcommand\tf{\ascii{TensorFlow}}\newcommand\quo[1]{“#1”}\newcommand\percent[1]{\ascii{#1\%}}\newcommand{\trans}{\emph{事务}}\newcommand{\act}{\emph{操作}}\newcommand{\seqact}{\emph{顺序操作}}\newcommand{\conact}{\emph{并发操作}}\newcommand{\atomact}{\emph{基本操作}}\newcommand{\syncact}{\emph{同步操作}}\newcommand{\asynact}{\emph{异步操作}}\newcommand{\action}[1]{\emph{\ascii{\itshape\_\_#1}}}\newcommand{\sigwait}{\action{sig\_wait}}\newcommand{\sigsync}{\action{sig\_sync}}\newcommand{\sigreply}{\action{sig\_reply}}\newcommand{\timerprot}{\action{timer\_prot}}\newcommand{\unknownevet}{\ascii{UNKNOWN\_EVENT}}\newcommand{\transdsl}{\ascii{Transaction DSL}}\newcommand{\oper}[1]{\ascii{Action#1}}\newcommand{\protproc}{\ascii{prot\_procedure}}%\newcommand{\code}[1]{\ascii{\small{\texttt{#1}}}}\newcommand{\code}[1]{\ascii{\footnotesize{\texttt{#1}}}}\newcommand{\script}[1]{\ascii{\scriptsize{\texttt{#1}}}}\newcommand{\inlinetitle}[1]{\large{\emph{#1}}}%\newcommand{\Email}{\begingroup \def\UrlLeft{<}\def\UrlRight{>} \urlstyle{tt}\Url}%\def\mailto|#1|{\href{mailto:#1}{Email|#1|}}\newcommand{\contrib}[2]{#1\quad{\small\quad\textit{#2}}\\[1ex]}\newcommand{\upcite}[1]{\textsuperscript{\cite{#1}}}
\newenvironment{enum}
{
\begin{spacing}{1.2}
% \begin{enumerate}[1.]
\begin{enumerate}
\setlength{\itemsep}{0pt}
\setlength{\itemindent}{2em}
%\setlength{\listparindent}{2em}
}{%
\end{enumerate}
\end{spacing}
}

\newenvironment{nitemize}
{
\begin{itemize}
\setlength{\itemsep}{0pt}
\setlength{\itemindent}{2em}
}{%
\end{itemize}
}


\newcommand{\suggest}[1]{
\tikzstyle{mybox} = [draw=black, very thick,
rectangle, rounded corners, inner sep=9pt, inner ysep=20pt]
\tikzstyle{fancytitle} =[fill=white, text=black, ellipse]
\noindent
\begin{tikzpicture}
\node [mybox] (box){%
\begin{minipage}{\textwidth}
\fangsong
#1
\end{minipage}
};
\node[fancytitle, right=10pt] at (box.north west) {\emph{建议}};
% \node[fancytitle, rounded corners] at (box.east) {$\clubsuit$};
\end{tikzpicture}
}

\newcommand{\notice}[1]{
\tikzstyle{mybox} = [draw=black, very thick,
rectangle, rounded corners, inner sep=9pt, inner ysep=20pt]
\tikzstyle{fancytitle} =[fill=white, text=black]
\noindent
\begin{tikzpicture}
\node [mybox] (box){%
\begin{minipage}{\textwidth}
\fangsong
#1
\end{minipage}
};
\node[fancytitle, right=10pt] at (box.north west) {\emph{注意}};
%\node[fancytitle, rounded corners] at (box.east) {$\clubsuit$};
\end{tikzpicture}
}

\newcommand{\tip}[1]{
\tikzstyle{mybox} = [draw=black, very thick,
rectangle, rounded corners, inner sep=9pt, inner ysep=20pt]
\tikzstyle{fancytitle} =[fill=white, text=black]
\noindent
\begin{tikzpicture}
\node [mybox] (box){%
\begin{minipage}{\textwidth}
\fangsong
#1
\end{minipage}
};
\node[fancytitle, right=10pt] at (box.north west) {\emph{提示}};
%\node[fancytitle, rounded corners] at (box.east) {$\clubsuit$};
\end{tikzpicture}
}

\newcommand\refch[1]{\ascii{第\ref{ch:#1}章(\nameref{ch:#1})}}
\newcommand\refsec[1]{\ascii{\ref{sec:#1}节(\nameref{sec:#1})}}

\newcommand\eitem[1]{\item {\itshape {#1}}}
\newcommand\cpp{\ascii{C\nobreak+\nobreak+}}
\newcommand\clang{\ascii{C}}
\newcommand\tf{\ascii{TensorFlow}}

\newcommand\quo[1]{“#1”}

\newcommand\percent[1]{\ascii{#1\%}}

\newcommand{\trans}{\emph{事务}}
\newcommand{\act}{\emph{操作}}
\newcommand{\seqact}{\emph{顺序操作}}
\newcommand{\conact}{\emph{并发操作}}
\newcommand{\atomact}{\emph{基本操作}}
\newcommand{\syncact}{\emph{同步操作}}
\newcommand{\asynact}{\emph{异步操作}}
\newcommand{\action}[1]{\emph{\ascii{\itshape\_\_#1}}}
\newcommand{\sigwait}{\action{sig\_wait}}
\newcommand{\sigsync}{\action{sig\_sync}}
\newcommand{\sigreply}{\action{sig\_reply}}
\newcommand{\timerprot}{\action{timer\_prot}}
\newcommand{\unknownevet}{\ascii{UNKNOWN\_EVENT}}
\newcommand{\transdsl}{\ascii{Transaction DSL}}
\newcommand{\oper}[1]{\ascii{Action#1}}
\newcommand{\protproc}{\ascii{prot\_procedure}}

%\newcommand{\code}[1]{\ascii{\small{\texttt{#1}}}}
\newcommand{\code}[1]{\ascii{\footnotesize{\texttt{#1}}}}
\newcommand{\script}[1]{\ascii{\scriptsize{\texttt{#1}}}}


\newcommand{\inlinetitle}[1]{\large{\emph{#1}}}

%\newcommand{\Email}{\begingroup \def\UrlLeft{<}\def\UrlRight{>} \urlstyle{tt}\Url}
%\def\mailto|#1|{\href{mailto:#1}{Email|#1|}}
\newcommand{\contrib}[2]{#1\quad{\small\quad\textit{#2}}\\[1ex]}


\newcommand{\upcite}[1]{\textsuperscript{\cite{#1}}}
Loading