\section{mlpack\+:\+:nn Namespace Reference}
\label{namespacemlpack_1_1nn}\index{mlpack\+::nn@{mlpack\+::nn}}
\subsection*{Classes}
\begin{DoxyCompactItemize}
\item 
class \textbf{ Sparse\+Autoencoder}
\begin{DoxyCompactList}\small\item\em A sparse autoencoder is a neural network whose aim to learn compressed representations of the data, typically for dimensionality reduction, with a constraint on the activity of the neurons in the network. \end{DoxyCompactList}\item 
class \textbf{ Sparse\+Autoencoder\+Function}
\begin{DoxyCompactList}\small\item\em This is a class for the sparse autoencoder objective function. \end{DoxyCompactList}\end{DoxyCompactItemize}
\subsection*{Functions}
\begin{DoxyCompactItemize}
\item 
void \textbf{ Maximal\+Inputs} (const arma\+::mat \&parameters, arma\+::mat \&output)
\begin{DoxyCompactList}\small\item\em Given a parameters matrix from an autoencoder, maximize the hidden units of the parameters, storing the maximal inputs in the given output matrix. \end{DoxyCompactList}\item 
void \textbf{ Normalize\+Col\+By\+Max} (const arma\+::mat \&input, arma\+::mat \&output)
\begin{DoxyCompactList}\small\item\em Normalize each column of the input matrix by its maximum value, if that maximum value is not zero. \end{DoxyCompactList}\end{DoxyCompactItemize}


\subsection{Function Documentation}
\mbox{\label{namespacemlpack_1_1nn_a5fcd73722265acc12d00ba8d32db6f17}} 
\index{mlpack\+::nn@{mlpack\+::nn}!Maximal\+Inputs@{Maximal\+Inputs}}
\index{Maximal\+Inputs@{Maximal\+Inputs}!mlpack\+::nn@{mlpack\+::nn}}
\subsubsection{Maximal\+Inputs()}
{\footnotesize\ttfamily void mlpack\+::nn\+::\+Maximal\+Inputs (\begin{DoxyParamCaption}\item[{const arma\+::mat \&}]{parameters,  }\item[{arma\+::mat \&}]{output }\end{DoxyParamCaption})}



Given a parameters matrix from an autoencoder, maximize the hidden units of the parameters, storing the maximal inputs in the given output matrix. 

Details can be found on the \textquotesingle{}Visualizing a Trained Autoencoder\textquotesingle{} page of the Stanford U\+F\+L\+DL tutorial\+:

{\tt http\+://deeplearning.\+stanford.\+edu/wiki/index.\+php/\+Main\+\_\+\+Page}

This function is based on the implementation (display\+\_\+network.\+m) from the \char`\"{}\+Exercise\+: Sparse Autoencoder\char`\"{} page of the U\+F\+L\+DL tutorial\+:

{\tt http\+://deeplearning.\+stanford.\+edu/wiki/index.\+php/\+Exercise\+:\+Sparse\+\_\+\+Autoencoder}

Example usage of this function can be seen below. Note that this function can work with the Columns\+To\+Blocks class in order to reshape the maximal inputs for visualization, as in the U\+F\+L\+DL tutorial. The code below demonstrates this.


\begin{DoxyCode}
arma::mat data; \textcolor{comment}{// Data matrix.}
\textcolor{keyword}{const} \textcolor{keywordtype}{size\_t} vSize = 64; \textcolor{comment}{// Size of visible layer, depends on the data.}
\textcolor{keyword}{const} \textcolor{keywordtype}{size\_t} hSize = 25; \textcolor{comment}{// Size of hidden layer, depends on requirements.}

\textcolor{keyword}{const} \textcolor{keywordtype}{size\_t} numBasis = 5; \textcolor{comment}{// Parameter required for L-BFGS algorithm.}
\textcolor{keyword}{const} \textcolor{keywordtype}{size\_t} numIterations = 100; \textcolor{comment}{// Maximum number of iterations.}

\textcolor{comment}{// Use an instantiated optimizer for the training.}
SparseAutoencoder<L\_BFGS> encoder(data, vSize, hSize);

arma::mat maximalInput; \textcolor{comment}{// Store the features learned by sparse autoencoder}
mlpack::nn::MaximalInputs(encoder.Parameters(), maximalInput);

arma::mat outputs;
\textcolor{keyword}{const} \textcolor{keywordtype}{bool} scale = \textcolor{keyword}{true};

ColumnsToBlocks ctb(5,5);
arma::mat output;
ctb.Transform(maximalInput, output);
\textcolor{comment}{// Save the output as PGM, for visualization.}
output.save(fileName, arma::pgm\_binary);
\end{DoxyCode}


\begin{DoxyPrecond}{Precondition}
Layout of parameters
\end{DoxyPrecond}
The layout of the parameters matrix should be same as following 
\begin{DoxyCode}
\textcolor{comment}{//          vSize   1}
\textcolor{comment}{//       |        |  |}
\textcolor{comment}{//  hSize|   w1   |b1|}
\textcolor{comment}{//       |\_\_\_\_\_\_\_\_|\_\_|}
\textcolor{comment}{//       |        |  |}
\textcolor{comment}{//  hSize|   w2'  |  |}
\textcolor{comment}{//       |\_\_\_\_\_\_\_\_|\_\_|}
\textcolor{comment}{//      1|   b2'  |  |}
\end{DoxyCode}


Also, the square root of v\+Size must be an integer (i.\+e. v\+Size must be a perfect square).


\begin{DoxyParams}{Parameters}
{\em parameters} & The parameters of the autoencoder. \\
\hline
{\em output} & Matrix to store the maximal inputs in. \\
\hline
\end{DoxyParams}
\mbox{\label{namespacemlpack_1_1nn_a9887e1e5ada8c7d5fb1ad59dc4d16b4a}} 
\index{mlpack\+::nn@{mlpack\+::nn}!Normalize\+Col\+By\+Max@{Normalize\+Col\+By\+Max}}
\index{Normalize\+Col\+By\+Max@{Normalize\+Col\+By\+Max}!mlpack\+::nn@{mlpack\+::nn}}
\subsubsection{Normalize\+Col\+By\+Max()}
{\footnotesize\ttfamily void mlpack\+::nn\+::\+Normalize\+Col\+By\+Max (\begin{DoxyParamCaption}\item[{const arma\+::mat \&}]{input,  }\item[{arma\+::mat \&}]{output }\end{DoxyParamCaption})}



Normalize each column of the input matrix by its maximum value, if that maximum value is not zero. 


\begin{DoxyParams}{Parameters}
{\em input} & The input data to normalize. \\
\hline
{\em output} & A matrix to store the input data in after normalization. \\
\hline
\end{DoxyParams}