[Tex/LaTex] Neural Network representation

algorithmsdiagramstikz-pgf

I would like to translate to Spanish a diagram that represents a neural network, the diagrama is something like this:
Step 1
Step 2

I would like to clone the design and effects, so I need some examples to do this,also I think that this image is something harder:
Step3

Best Answer

I went ahead and used Fernando Martinez's example to get me started. This uses LuaLaTex:

\RequirePackage[dvipsnames]{xcolor}
\documentclass[tikz]{standalone}
\usetikzlibrary{calc,arrows}
\begin{document}
\begin{tikzpicture}

%%Create a style for the arrows we are using
\tikzset{normal arrow/.style={draw,-triangle 45}}

%%Create the different coordinates to place layer 1 nodes
\path (0,0) coordinate (l1n1) ++(0,-2) coordinate (l1n2) ++(0,-2) coordinate (l1n3) ++(0,-2) coordinate (l1n4) ++(0,-2) coordinate (l1n5) ++(0,-2) coordinate (l1n6) ++(0,-2) coordinate (l1n7) ++(0,-2) coordinate (l1n8) ++(0,-2) coordinate (l1n9);

%%Create the different coordinates to place INPUT layer (xs)
\foreach \i in {1,2,3,4,5,6,7,8,9}{ \path (l1n\i) ++(-5,1) coordinate (x\i); }

%%Create the different coordinates to place Outputs
\foreach \i in {1,2,3,4,5,6,7}{ \path (l1n\i) ++(8,-1) coordinate (o\i); }

%%generate the second level top node points
\path ($(l1n1)!.5!(l1n2)!5 cm!90:(l1n2)$) coordinate (l2n0);

%%Create the different coordinates to place second layer nodes
\path (l2n0) ++(0,-2) coordinate (l2n1) ++(0,-2) coordinate (l2n2) ++(0,-2) coordinate (l2n3) ++(0,-2) coordinate (l2n4) ++(0,-2) coordinate (l2n5) ++(0,-2) coordinate (l2n6) ++(0,-2) coordinate (l2n7);

%%generate the position of last second level node point
\path ($(l1n5)!.5!(l1n6)!3 cm!90:(l1n6)$) coordinate (l2nx);


%%Place nodes
\foreach \i in {0,1,2,3,4,5,6}{
    \node[draw,circle] (cl2n\i) at (l2n\i) {\phantom{a}$\sigma_\i\phantom{a}$};
}
\foreach \i in {1,2,3,4,5,6,7,8}{
    \node[draw,circle] (cl1n\i) at (l1n\i) {\phantom{a}$\sigma_{\directlua{tex.sprint(\i + 6)}}\phantom{a}$};
}

%%Label output nodes
\node (lo1) at (o1) {build wind float};
\node (lo2) at (o2) {build wind non-float};
\node (lo2) at (o3) {vehic wind float};
\node (lo3) at (o4) {vehic wind non-float};
\node (lo4) at (o5) {containers};
\node (lo5) at (o6) {tableware};
\node (lo6) at (o7) {headlamps};

%%Label input nodes
\node (nx1) at (x1) {$RI$};
\node (nx2) at (x2) {$Na$};
\node (nx3) at (x3) {$Mg$};
\node (nx4) at (x4) {$Al$};
\node (nx5) at (x5) {$Si$}; 
\node (nx6) at (x6) {$K$};
\node (nx7) at (x7) {$Ca$};
\node (nx8) at (x8) {$Ba$};
\node (nx9) at (x9) {$Fe$};

%%Drawing arrows 
%%Transparent arrows between input and Layer1
\foreach \i in {1,2,3,4,5,6,7,8,9}{
    \foreach \j in {1,2,3,4,5,6,7,8}{
        \path[normal arrow, Cyan, draw opacity=0.2] (nx\i) -- (cl1n\j);
    }
}

%%Explicit arrows between Al and Layer1
\foreach \i/\val in {1/4.8,2/-0.14,3/-6.75,4/4.07,5/-2.88,6/14.18,7/-12.79,8/-2.41}{ 
    \newcommand\dosomecoolmath{\directlua{ x = 0.862679-(\i)*(-0.000529583) + ((-\i)^(2))*(-0.14934524) tex.sprint(x)}}
    \path[normal arrow, draw opacity=0.5] (nx4) -- node[above=\dosomecoolmath em] {$\mathbf{w_{(x4)\directlua{tex.sprint(\i + 6)}} = \val}$} (cl1n\i);
}

%%Explicit arrows between Layer1 and Layer2
\foreach \i/\val in {1/-0.199, 2/2.1155,3/-2.921,4/-7.29,5/-3.38,6/6.724,7/-2.56,8/4.470}{ 
    \newcommand\dosomecoolmath{
        \directlua{x = 8.862679-(\i)*(-0.000529583) + ((-\i)^(2))*(-0.22934524)
        tex.sprint(x)}}
        \path[normal arrow, draw opacity=0.7] (cl1n\i) -- node[above=\dosomecoolmath em] {$\mathbf{w_{(\directlua{tex.sprint(\i + 6)})4} = \val}$} (cl2n4);
        \foreach \j in {0,1,2,3,5,6}{
            \path[normal arrow, Cyan, draw opacity=0.2] (cl1n\i) -- (cl2n\j);
        }
}

%Draw final threshold
\path (o5) ++(0,-.5) coordinate (thres); \node (threshold) at (thres) {\fbox{$-3.617302530741029$}};
\end{tikzpicture}
\end{document}

Which produces: Results

Related Solutions

[Tex/LaTex] drawing back propagation neural network

I was not sure about the desired position of the arrow and the text, so I used

\draw[->] (mat-6-3) -- ++(0pt,3cm) -| node[pos=0.15,above] {Error back propagation} ( $ (mat-2-1)!0.5!(mat-2-2) $ );

but you can change those settings according to your needs. The code:

\documentclass{article}
\usepackage{tikz}
\usetikzlibrary{matrix,chains,positioning,decorations.pathreplacing,arrows,calc}

\tikzset{
block/.style={
  draw,
  rectangle, 
  text width=3em, 
  text centered, 
  minimum height=8mm,     
  node distance=2.3em
  }, 
line/.style={draw}
}

\begin{document}
    \begin{tikzpicture}[
    plain/.style={
      draw=none,
      fill=none,
      },
    net/.style={
      matrix of nodes,
     nodes={
       draw,
        circle,
    inner sep=10pt
    },
  nodes in empty cells,
  column sep=2cm,
  row sep=-9pt
  }, 
 >=latex
 ]
\matrix[net] (mat)
{ 
|[plain]| \parbox{1cm}{\centering Input\\layer} & |[plain]| \parbox{1cm}{\centering     Hidden\\layer} & |[plain]| \parbox{1cm}{\centering Output\\layer} \\
& |[plain]| \\
|[plain]| & \\
& |[plain]| \\
|[plain]| & |[plain]| \\
& & \\
|[plain]| & |[plain]| \\
& |[plain]| \\
|[plain]| & \\
& |[plain]| \\
};
\foreach \ai [count=\mi ]in {2,4,...,10}
  \draw[<-] (mat-\ai-1) -- node[above] {Input \mi} +(-2cm,0);
\foreach \ai in {2,4,...,10}
{\foreach \aii in {3,6,9}
  \draw[->] (mat-\ai-1) -- (mat-\aii-2);
}
\foreach \ai in {3,6,9}
  \draw[->] (mat-\ai-2) -- (mat-6-3);
%\draw[->] (mat-6-3) -- node[above] {Ouput} +(2cm,0);
\path [line] node{error} -- (mat-1-1);
\draw[->] (mat-6-3) -- ++(0pt,3cm) -| node[pos=0.15,above] {Error back propagation} ( $ (mat-2-1)!0.5!(mat-2-2) $ );
\end{tikzpicture}

\end{document}

enter image description here

[Tex/LaTex] Bottom up neural network with TikZ

Update: The OP asked a second question. This solution modifies the first solution and removed some redundancy, hoping this time it compiles.

enter image description here

Code

\documentclass{article}

\usepackage{tikz}
\usetikzlibrary{calc}

\begin{document}
\pagestyle{empty}

\def\layersep{2.5cm}

\tikzset{neuron/.style={circle,thick,fill=black!25,minimum size=17pt,inner sep=0pt},
    input neuron/.style={neuron, draw,thick, fill=gray!30},
    hidden neuron/.style={neuron,fill=white,draw},
    hoz/.style={rotate=-90}}   %<--- for labels

\begin{tikzpicture}[-,draw=black, node distance=\layersep,transform shape,rotate=90]  %<-- rotate the NN

% Draw the input layer nodes
\foreach \name / \y in {1/1,2/2,3/3,5/m}
\node[input neuron, hoz] (I-\name) at (0,-\name) {\color{red}$b_\y$};

\node[hoz] (I-4) at (0,-4) {$\dots$};

\foreach \name / \y in {1/1,2/2,3/3,5/m}
\path[hoz] (I-\name) node[below=0.5cm](0,-\name) {$v_\y$};

% Draw the hidden layer nodes
\foreach \name / \y in {1/1,2/2,3/3,6/n}
\path[yshift=0.5cm] node [hidden neuron, hoz] (H-\name) at (\layersep,-\name cm) {\color{red}$c_\y$};

\path[yshift=0.5cm]
   node[hoz] () at (\layersep,-4 cm) {$\dots$};
\path[yshift=0.5cm]
   node[hoz] () at (\layersep,-5 cm) {$\dots$};

\foreach \name / \y in {1/1,2/2,3/3,6/n}
\path[hoz] (H-\name) node[above=0.5cm]  {$h_\y$};  

\path node[hoz,right] at ($(I-5)!0.5!(H-6)$) {\color{red}$w_{nm}$};       

% Connect every node in the input layer with every node in the  hidden layer.
    \foreach \source in {1,2,3,5}
        \foreach \dest in {1,2,3,6}
            \path (I-\source.north) edge (H-\dest.south);
\end{tikzpicture}
% End of code
\end{document}

-------------------------- first edition

This is one possibility, draw as usual, then rotate the tikzpicture with tansform shape option and label respectively after completed, as shown below,

enter image description here

Code

\documentclass{article}

\usepackage{tikz}
\begin{document}
\pagestyle{empty}

\def\layersep{2.5cm}

\begin{tikzpicture}[shorten >=1pt,->,draw=black!50, node distance=\layersep,transform shape,rotate=90]  %<-- rotate the NN
    \tikzstyle{every pin edge}=[<-,shorten <=1pt]
    \tikzstyle{neuron}=[circle,fill=black!25,minimum size=17pt,inner sep=0pt]
    \tikzstyle{input neuron}=[neuron, fill=green!50];
    \tikzstyle{output neuron}=[neuron, fill=red!50];
    \tikzstyle{hidden neuron}=[neuron, fill=blue!50];
    \tikzstyle{annot} = [text width=4em, text centered]
    \tikzset{hoz/.style={rotate=-90}}   %<--- for labels
    % Draw the input layer nodes
    \foreach \name / \y in {1,...,4}
    % This is the same as writing \foreach \name / \y in {1/1,2/2,3/3,4/4}
        \node[input neuron, pin=left:\rotatebox{-90}{\parbox[t][][r]{8mm}{\centering Input \\\#\y}}] (I-\name) at (0,-\y) {};

    % Draw the hidden layer nodes
    \foreach \name / \y in {1,...,5}
        \path[yshift=0.5cm]
            node[hidden neuron] (H-\name) at (\layersep,-\y cm) {};

    % Draw the output layer node
    \node[output neuron,pin={[pin edge={->}]right:\rotatebox{-90}{Output}}, right of=H-3] (O) {};

    % Connect every node in the input layer with every node in the
    % hidden layer.
    \foreach \source in {1,...,4}
        \foreach \dest in {1,...,5}
            \path (I-\source) edge (H-\dest);

    % Connect every node in the hidden layer with the output layer
    \foreach \source in {1,...,5}
        \path (H-\source) edge (O);

    % Annotate the layers
    \node[annot,above of=H-1, node distance=1cm,hoz] (hl) {Hidden layer};
    \node[annot,left of=hl,hoz] {Input layer};
    \node[annot,right of=hl,hoz] {Output layer};
\end{tikzpicture}
% End of code
\end{document}

Best Answer

Related Solutions

[Tex/LaTex] drawing back propagation neural network

[Tex/LaTex] Bottom up neural network with TikZ

Related Question