TensorFlow.tex

\documentclass[12pt]{article}


\usepackage{amssymb,amsmath,amsthm}

\usepackage{wrapfig,tikz, tikz-cd}
\usetikzlibrary{arrows, arrows, calc, decorations.markings, automata,calc}

\newcommand{\dd}{\partial}
\newcommand{\argmax}{\operatorname{argmax}}
\newcommand{\cut}{\operatorname{cut}}
\newcommand{\Span}{\operatorname{Span}}
\newcommand{\RR}{\mathbb{R}}
\newcommand{\KK}{\mathbb{K}}
\newcommand{\rr}{\text{r}}
\newcommand{\al}{\alpha}
\newcommand{\be}{\beta}
\newcommand{\eps}{\varepsilon}
\newcommand{\ph}{\varphi}
\newcommand{\ZZ}{\mathbb{Z}}
\newcommand{\BB}{\mathfrak{B}}
\newcommand{\MM}{\mathfrak{M}}
\newcommand{\kk}{\textbf{k}}

\usepackage{hyperref}
\hypersetup{
    colorlinks=true,
    linkcolor=blue,
    filecolor=magenta,
    urlcolor=cyan,
}

\usepackage{tcolorbox}

\definecolor{tangoBlack1}{RGB}{0,0,0}
\definecolor{tangoGrey1}{RGB}{210,210,210}

\makeatletter
\newtcbox{\keywordmin}{
  on line,
  fontupper=\sffamily,
  boxrule=1pt,
  arc=1pt,
  coltext=tangoBlack1,
  colback=tangoGrey1,
  colframe=tangoGrey1,
  boxsep=0pt,
  left=2pt,right=2pt,top=2pt,bottom=2pt
}
\makeatother

\begin{document}



\title{Tensor Flow cheat sheet}
\author{Efim Abrikosov}
\maketitle

\section{Data Prepocessing}

\subsection{Feature columns}
\href{https://www.tensorflow.org/guide/feature_columns}{\keywordmin{tf.feature\_column}}

\begin{itemize}
  \item \keywordmin{tf.feature\_column.numeric\_column(\emph{colname})}
  \item\keywordmin{tf.feature\_column.bucketized\_column(\emph{tf.feature\_coumn.numeric\_column}, \emph{boundaries\_list})}
  \item\keywordmin{tf.feature\_column.crossed\_column(\emph{[tf.feature\_coumn.categorical\_column list]}, \emph{num\_hash\_buckets})}
  \item\keywordmin{tf.feature\_column.categorical\_column\_with\_identity(\emph{colname}, \emph{num\_buckets})}
  \item\keywordmin{tf.feature\_column.categorical\_column\_with\_vocabulary\_list(\emph{colname}, \emph{vocabulary\_list})}
  \item\keywordmin{tf.feature\_column.categorical\_column\_with\_hash\_bucket()\emph{colname}, \emph{hash\_bucket\_size})}
  \item\keywordmin{tf.feature\_column.embedding\_column(\emph{tf.feature\_column.categorical\_column}, \emph{n\_dim})}
\end{itemize}

\section{Estimator API}
\href{https://www.tensorflow.org/guide/estimators}{\keywordmin{tf.estimator}}
\begin{itemize}
  \item \keywordmin{tf.estimator.Estimator(\emph{model\_fn, out\_dir})}
  \begin{itemize}
    \item \emph{model\_fn} is a function that returns \emph{tf.estimator.EstimatorSpec} given a triple (\emph{feature\_cols}, \emph{targets}, \emph{mode})

    Sample signature: \keywordmin{tf.estimator.EstimatorSpec(\emph{**args})}

    Arguments list consists of:
    \begin{enumerate}
      \item \emph{mode}, one of the three:

       \keywordmin{tf.estimator.ModeKeys.TRAIN},

        \keywordmin{tf.estimator.ModeKeys.EVAL},

        \keywordmin{tf.estimator.ModeKeys.PREDICT}
      \item \emph{predictions\_dict}
      \item \emph{loss}
      \item \emph{train\_op}
      \item \emph{eval\_metrics\_ops}
      \item \emph{export\_outputs}
    \end{enumerate}
    \item \emph{model\_dir} is the name of output directory (string)
  \end{itemize}
  \item \keywordmin{tf.estimator.train\_and\_evaluate(estimator, train\_spec, eval\_spec)}
    \begin{itemize}
      \item \emph{estimator} is an instance of \emph{tf.estimator}

      Sample singature: \keywordmin{tf.estimator.\emph{[model\_name]}(\emph{model\_dir}, \emph{feature\_columns}, \emph{model\_params})}
      \item \emph{train\_spec} is an instance of \emph{tf.estimator.TrainSpec}

      Sample signature: \keywordmin{tf.estimator.TrainSpec(\emph{input\_fn}, \emph{max\_steps})}
      \item \emph{eval\_spec} is an instance of \emph{tf.estimator.EvalSpec}

      Sample signature: \keywordmin{tf.estimator.EvalSpec(\emph{input\_fn},\emph{steps}, \emph{start\_delay\_secs}, \emph{throttle\_secs})}
    \end{itemize}
    \item Adding performance metrics

    Sample signature:
    \keywordmin{\emph{estimator} = tf.contrib.estimator.add\_metrics(\emph{estimator},\emph{rmse})}
\end{itemize}

\section{TensorBoard}

\begin{itemize}
  \item \keywordmin{\emph{filewriter\_name} = tf.summary.FileWriter(\emph{path\_name}, \emph{tf.graph} = tf.get\_default\_graph())}
  \item \keywordmin{\emph{filewriter\_name}.close()}
  \item in terminal or with \%\%bash magic
  \begin{itemize}
    \item \$ tensorboard -{}-logdir \emph{path\_name}
  \end{itemize}

\end{itemize}



\end{document}