April 05, 2024
When creating plots of data, sometimes there is the need to zoom in on certain sections of the plot.
Tikz offers the spy library, which shows part of a tikzpicture.
However, this is a so-called canvas transformation, which enlarges everything.
While this often works fine for regular figures, in my experience it usually fails when applied to plots of data, as the line widths etc. are also scaled, which makes it hard to read the plot.
A better way to achieve zoom-ins in plots is to create an entire new plot and superimpose it on the existing picture. This is actually easy to achieve and offers many options for fine-control.
As an example, here is what we will be creating (taken from my 2023 ISCAS paper):
First, let's set up the basic plot (sample data file):
\documentclass{standalone}
\usepackage{pgfplots}
\pgfplotsset{compat = newest}
\begin{document}
\begin{tikzpicture}
\begin{axis}
[
grid=both,
no markers,
line join = round,
enlarge x limits = false,
enlarge y limits = false,
width=7cm,
height=3cm,
scale only axis,
ymin = 0,
ymax = 0.8,
]
\addplot table[col sep = comma]
{pgfplots_zoom_sample_data.csv};
\end{axis}
\end{tikzpicture}
\end{document}This creates the following plot (for simplicity I've removed all irrelevant stuff, which makes the plot nicer but the code longer):
If the spy tikzlibrary is used, the result is something like the following (in reality, the SVG looks messed up, because the entire plot of the spy is visible. I've post-edited the file to remove this, so it's not exactly like it should look):
The code for the above image is:
\documentclass{standalone}
\usepackage{tikz, pgfplots}
\usetikzlibrary{spy}
\pgfplotsset{compat = newest}
\begin{document}
\begin{tikzpicture}
\def\zoomxsource{1}
\def\zoomysource{0.35}
\def\zoomxtarget{1}
\def\zoomytarget{0.65}
[
spy using outlines = {
circle,
magnification = 6,
size = 1cm,
connect spies
},
]
\begin{axis}
[
grid=both,
no markers,
line join = round,
enlarge x limits = false,
enlarge y limits = false,
width=7cm,
height=3cm,
scale only axis,
each nth point = 10,
filter discard warning = false,
unbounded coords = discard,
ymin = 0,
ymax = 0.8,
]
\addplot table[col sep = comma] {data_full.csv};
\coordinate (zoomtarget)
at (axis cs: \zoomxtarget, \zoomytarget);
\coordinate (zoomsource)
at (axis cs: \zoomxsource, \zoomysource);
\end{axis}
\spy on (zoomsource) in node at (zoomtarget);
\end{tikzpicture}
\end{document}
I used \def to define the locations of the zoom source and target.
This is not required of course, but helps making the semantics of the spy a bit clearer.
Furthermore, this approach will also be used later on, where we will have to do a few things more manually, so keeping a clear code structure is good practice.
While the above result does not look too bad, it has a few issues:
First, the line widths of the curve and the grid are enlarged.
While this might be the right choice in rare occasions, I think in plots this is mostly unwanted.
Furthermore (this is a bit more subtle and depends on the data) there is no control of the plot settings for the spy.
In this particular example, in the paper the curve of the large plot is drawn with fewer points for efficiency (also, it makes the curve a bit nicer).
But in the detailed plot (the spy), all the points should be used.
Stuff like this is simply impossible with the spy library.
Therefore, a better method is required.
For this, we are going to create another plot on top.
In a tikzpicture we can use multiple axis environments, which we will use here.
In the subsequent code examples, I will only show the code inside the tikzpicture, the preamble stays the same.
The structure of the plotting code will be like this:
\begin{tikzpicture}
% (1) \def's for the zoom source and target
\begin{axis}
% main plot
% (2) node definitions for the zoomed plot
\end{axis}
\begin{axis}
[
at = {(zoomtarget)}
]
% (3) zoomed-in plot
\end{axis}
\end{tikzpicture}
The definitions for the source and the target concern the x- and y-coordinates (in axis coordinates).
The source is the point of the plot that should be zoomed in on (the 'on' part of the tikz spy).
The target is the point where the magnified plot will reside (the 'at' part of the tikz spy).
In this case, we want to zoom in on the curve at (1, 0.35).
The size of the zoom also concerns the size in axis coordinates.
The magnification factor (as in the tikz spy) is only present implicitly by the size of the magnifying plot.
% node definitions for the zoomed plot
\def\zoomxsource{1}
\def\zoomysource{0.35}
\def\zoomxsize{0.04}
\def\zoomysize{0.04}
\def\zoomxtarget{1}
\def\zoomytarget{0.35}
\def\zoomtargetwidth{1cm}
\def\zoomtargetheight{0.6cm}
With the definitions of the source and the target coordinates and the sizes, tikz nodes can now be created which will be used for placing the magnifying plot.
The creation of nodes utilizes the axis cs coordinate system.
This ensures that the nodes can be created by using coordinates from the curves, not from internal tikz coordinate systems.
In newer pgfplots versions this is the default, using axis cs makes sure that this also works with older compatibility settings/versions.
Furthermore, the node coordinates use the ({x, y}) notation, which enables simple calculations.
This is used to calculate the corner points of the magnifying rectangle.
% (2) node definitions for the zoomed plot
\coordinate (zoomtarget) at
(axis cs: \zoomxtarget, \zoomytarget);
\coordinate (zoomsourcebl) at
({axis cs:
\zoomxsource - \zoomxsize / 2,
\zoomysource - \zoomysize / 2
});
\coordinate (zoomsourcetr) at
({axis cs:
\zoomxsource + \zoomxsize / 2,
\zoomysource + \zoomysize / 2
});Now to the creation of the magnified plot (3). Essentielly, this is almost the same as the main plot, but leaves out the axes and some other stuff and it is placed at the target node. Therefore, the skeleton for the plot is:
% (3) zoomed-in plot
\begin{axis}
[
anchor = center,
at = {(zoomtarget)},
scale only axis,
width = \zoomtargetwidth,
height = \zoomtargetheight,
axis lines = none,
xmin = \zoomxmin,
xmin = \zoomxmax,
clip mode = individual
]
% \addplot just like in the main axis environment
\end{axis}Let's go through the options one-by-one. These are the essentiell ones, of course more can be used to fine-tune the appearance.
anchor = center and at = {(zoomtarget)} places the plot at this coordinate. The extra braces and parantheses are required.
scale only axis makes sure that the given size (width and height) does not take into account any labels, titles, tick labels etc.
axis lines = none gets rid of any axis lines (duh).
I guess it is also possible to draw a magnified part of the picture with axis lines, but I don't think this is what most people expect and want.
xmin and xmax restrict the plot to only the part that we want to show.
Lastly, clip mode = individual makes sure that only plot paths are clipped inside of the (non-visible) axes.
This enables us to place nodes for annotations. This option is not strictly necessary, but as I'm placing annotations it makes sense to use.
The final result looks like this:
This concludes this tutorial. The full example is a bit lengthy for this blog and is therefore available as file. For convenience, here is the link for the used data file.