Neural Network illustration

Question

I'm trying to create a neural network illustration using the neuralnetwork package. I want to label my bias-neurons separately.

Right now I have the code:

\documentclass[fleqn,11pt,a4paper,final]{article} 
\usepackage{neuralnetwork}
\begin{document}

\begin{figure}[h!]
\begin{neuralnetwork}[height=1.5 ,maintitleheight=1cm,layertitleheight=3cm]
  \newcommand{\nodetextx}[2]{$x_{#2}$}
  \newcommand{\nodetextz}[2]{$y$}
  \newcommand{\nodetexth}[2]{$h_{#2,1}$}
  \newcommand{\nodetexthto}[2]{$h_{#2,2}$}
  \inputlayer[count=3, title={Input-lag}, text=\nodetextx]
  \hiddenlayer[count=5, title={Skjult lag}, text=\nodetexth]
  \linklayers
  \hiddenlayer[count=4, title={Skjult lag}, text=\nodetexthto]
   \linklayers
  \outputlayer[count=1, title={Output-lag}, text=\nodetextz] 
  \linklayers
\end{neuralnetwork}
\end{figure}
\end{document}

Which gives me the following illustration:

I want the x_0-neuron to be labelled \textbf{c_1}, the h_{0,1}-neuron to be labelled \textbf{c_2} and finally the h_{0,2}-neuron to be labelled b. How do I do this?

Help is appreciated. Thanks

score 9 · Accepted Answer · answered Jul 06 '17 at 11:05

This might be more complicated than it needs be, but it does seem to work.

\documentclass[border=5mm]{standalone}
\usepackage{xpatch}
\usepackage{neuralnetwork}
% this is the command that prints the first node in a layer
%    \node[bias neuron] (L\nn@layerindex-0) at (\nn@node@xb, \nn@node@y) {\nn@nodecaption{\nn@layerindex}{0}};
% we want to modify the last bit to use the loop macro \nn@nodeindex instead of 0
\makeatletter
\xpatchcmd{\layer}{\nn@nodecaption{\nn@layerindex}{0}}{\nn@nodecaption{\nn@layerindex}{\nn@nodeindex}}{}{}

% define the nodeindex to be zero initially
\newcommand\nn@nodeindex{0}

\newcommand{\nodetextx}[2]{
  \ifnum \nn@nodeindex=0
     $\mathbf{c_1}$
  \else
     $x_{#2}$
  \fi
}

\newcommand{\nodetextz}[2]{$y$}

\newcommand{\nodetexth}[2]{
  \ifnum \nn@nodeindex=0
    $\mathbf{c_2}$
  \else
    $h_{#2,1}$
  \fi
}

\newcommand{\nodetexthto}[2]{
  \ifnum \nn@nodeindex=0
    $b$
  \else
    $h_{#2,2}$
  \fi
}
\makeatother

\begin{document}
\begin{neuralnetwork}[height=1.5 ,maintitleheight=1cm,layertitleheight=3cm]
  \inputlayer[count=3, title={Input-lag}, text=\nodetextx]
  \hiddenlayer[count=5, title={Skjult lag}, text=\nodetexth]
  \linklayers
  \hiddenlayer[count=4, title={Skjult lag}, text=\nodetexthto]
  \linklayers
  \outputlayer[count=1, title={Output-lag}, text=\nodetextz] 
  \linklayers
\end{neuralnetwork}
\end{document}

score -1 · Answer 2 · edited Jun 27 '18 at 06:14

to resolve

I want to contribute to my solution. Get resolved by putting the bias = false tag. I do not know what bias is for anymore I found it on another blog and it worked.

\documentclass[border=5mm]{standalone}
\usepackage{xpatch}
\usepackage{neuralnetwork}
  % this is the command that prints the first node in a layer
%    \node[bias neuron] (L\nn@layerindex-0) at (\nn@node@xb, \nn@node@y) {\nn@nodecaption{\nn@layerindex}{0}};
% we want to modify the last bit to use the loop macro \nn@nodeindex instead of 0
\makeatletter
\xpatchcmd{\layer}{\nn@nodecaption{\nn@layerindex}{0}}{\nn@nodecaption{\nn@layerindex}{\nn@nodeindex}}{}{}

% define the nodeindex to be zero initially
\newcommand\nn@nodeindex{0}

\newcommand{\nodetextx}[2]{
  \ifnum \nn@nodeindex=0
     $\mathbf{c_1}$
  \else
     $i_{#2}$
  \fi
}

\newcommand{\nodetextz}[2]{$o_1$}

\newcommand{\nodetexth}[2]{
  \ifnum \nn@nodeindex=0
    $\mathbf{c_2}$
  \else
    $h_{1, #2}$
  \fi
}

\newcommand{\nodetexthto}[2]{
  \ifnum \nn@nodeindex=0
    $b$
  \else
    $h_{2, #2}$
  \fi
}
\makeatother

\begin{document}
\begin{neuralnetwork}[height=1.5 ,maintitleheight=1cm,layertitleheight=5cm]
  \inputlayer[count=5, bias=false, title={Input}, text=\nodetextx]

  \hiddenlayer[count=10, bias=false, title={Oc1}, text=\nodetexth]
  \linklayers
  \hiddenlayer[count=10, bias=false, title={Oc2}, text=\nodetexthto]
  \linklayers

  \outputlayer[count=1, title={Output}, text=\nodetextz] 
  \linklayers
\end{neuralnetwork}
\end{document}

I guess your point is that you've eliminated the bias nodes. But, since labelling those is the point of the question, eliminating them is hardly an answer to it. Am I missing something? You seem to be answering some other question and your answer may be a good answer to that question, but it doesn't seem to be any answer at all to this question. — cfr, Jun 27 '18 at 03:37

Neural Network illustration

2 Answers2

Linked