Problem with scaling entire tikzpicture with transform canvas

Question

I read this. They suggest using transform canvas, which is rescaling but also it moves the picture where there is text (see figure). I also tried using \resizebox{}{} or \scalebox{}{}, bu that just makes the picture disappear.

This is how I generate the picture

\begin{center}
\begin{tikzpicture}[thick,scale=1.2,transform canvas={scale=1.2}, 
     clear/.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
]
% define matrix mat to hold nodes
% using net as default style for cells
\matrix[net] (mat)
{
% Define layer headings
|[clear]| \parbox{1.3cm}{\centering Input\\layer} 
    & |[clear]| \parbox{1.3cm}{\centering Hidden\\layer} 
    & |[clear]| \parbox{1.3cm}{\centering Output\\layer} \\
|[clear]| &      & |[clear]| \\         
  &     |[clear]|   & |[clear]| \\
  |[clear]| &   &|[clear]| \\
  & |[clear]| & |[clear]| \\
  |[clear]| & & \\
  & |[clear]| & |[clear]| \\
  |[clear]| & & |[clear]| \\
   & |[clear]| & |[clear]| \\
   |[clear]| & & |[clear]| \\
};


\draw[-] (mat-3-1) -- node[above] {$s_t^1$} +(-1.5cm,0);
\draw[-] (mat-5-1) -- node[above] {$s_t^2$} +(-1.5cm,0);
\draw[-] (mat-7-1) -- node[above] {$s_t^3$} +(-1.5cm,0);
\draw[-] (mat-9-1) -- node[above] {$s_t^4$} +(-1.5cm,0);
% lines from a_{i}^{0} to each a_{j}^{1}
\foreach \ai in {3,5,...,9} {
    \foreach \aii in {2,4,6,8,10}
        \draw[-] (mat-\ai-1) -- (mat-\aii-2);
        }
% lines from a_{i}^{1} to a_{0}^{2}
\foreach \ai in {2,4,6,8,10}
  \draw[-] (mat-\ai-2) -- (mat-6-3);

% right most line with Output label
\draw[->] (mat-6-3) -- node[above] {Output} +(2cm,0);
\end{tikzpicture}
\end{center}

Does anyone have any ideas how to fix that?

Answer 1

One way to be able to use \scalebox and the like here is to add an ampersand replacement to the matrix.

\documentclass{article}
\usepackage{tikz}
\usetikzlibrary{matrix}
\begin{document}
\begin{center}
\scalebox{0.5}{\begin{tikzpicture}[thick,
     clear/.style={ 
         draw=none,
         fill=none
     },
     net/.style={ampersand replacement=\&,
         matrix of nodes,
         nodes={ draw, circle, inner sep=10pt },
         nodes in empty cells,
         column sep=2cm,
         row sep=-9pt
     },
     >=latex
]
% define matrix mat to hold nodes
% using net as default style for cells
\matrix[net] (mat)
{
% Define layer headings
|[clear]| \parbox{1.3cm}{\centering Input\\layer} 
    \& |[clear]| \parbox{1.3cm}{\centering Hidden\\layer} 
    \& |[clear]| \parbox{1.3cm}{\centering Output\\layer} \\
|[clear]| \&      \& |[clear]| \\         
  \&     |[clear]|   \& |[clear]| \\
  |[clear]| \&   \&|[clear]| \\
  \& |[clear]| \& |[clear]| \\
  |[clear]| \& \& \\
  \& |[clear]| \& |[clear]| \\
  |[clear]| \& \& |[clear]| \\
   \& |[clear]| \& |[clear]| \\
   |[clear]| \& \& |[clear]| \\
};
\draw[-] (mat-3-1) -- node[above] {$s_t^1$} +(-1.5cm,0);
\draw[-] (mat-5-1) -- node[above] {$s_t^2$} +(-1.5cm,0);
\draw[-] (mat-7-1) -- node[above] {$s_t^3$} +(-1.5cm,0);
\draw[-] (mat-9-1) -- node[above] {$s_t^4$} +(-1.5cm,0);
% lines from a_{i}^{0} to each a_{j}^{1}
\foreach \ai in {3,5,...,9} {
    \foreach \aii in {2,4,6,8,10}
        \draw[-] (mat-\ai-1) -- (mat-\aii-2);
        }
% lines from a_{i}^{1} to a_{0}^{2}
\foreach \ai in {2,4,6,8,10}
  \draw[-] (mat-\ai-2) -- (mat-6-3);
% right most line with Output label
\draw[->] (mat-6-3) -- node[above] {Output} +(2cm,0);
\end{tikzpicture}}
\end{center}
\end{document}

One should say that this cures the symptoms. Yet to have a more elegant solution you may want to (fundamentally) rewrite the code, starting from questioning the use of the \parboxes in a matrix.

One way of getting a very similar output is

\documentclass{article}
\usepackage{tikz}
\usetikzlibrary{positioning}
\begin{document}
\begin{center}
\begin{tikzpicture}[scale=0.8,transform shape,
neuron/.style={circle,inner sep=1em,draw}]
 \begin{scope}[x=3cm,y=1.5cm,local bounding box=net]  
 \foreach \Y [count=\Z,remember=\Y as \LastY] in 
 {4,5,1} 
 {\foreach \YY in {1,...,\Y}
 {\node[neuron] (n-\Z-\YY) at (\Z,-\YY+\Y/2+1/2) {};
 \ifnum\Z=1
  \draw (n-\Z-\YY.west) -- node[above] {$s^t_{\YY}$}++ (-2em,0);
 \else
  \foreach \YYY in {1,...,\LastY}
  {\draw[-,semithick] (n-\the\numexpr\Z-1\relax-\YYY) -- (n-\Z-\YY) ;}
 \fi
 }}
 \end{scope} 
 \path foreach \X [count=\Y] in {Input,Hidden,Outpt}
 { (net.north-|n-\Y-1) node[above=1em,align=center] {\X\\ Layer}
 };
\end{tikzpicture}
\end{center}
\end{document}

I put a nontrivial scale factor in to show that it works. However, you have other, arguably better handles to control the size of this diagram. By changing x and y in the scope you control the horizontal and vertical distances between the neurons, respectively. The sizes of the latter are controlled by the style. This code also avoids the need for adding |[clear]| all over, and if you want to change the number of neurons per layer, all you need to change is {4,5,1}. Of course, there are many other examples of drawing neuronal networks on this site (and I thought there is even a package but for some reason I can't find it now).