Given its scientific roots (that means several different needs), Matplotlib has a lot of output formats available, which can be used for articles/books and other print publications, for web pages, or for any other reason we can think of. Let's first differentiate the output formats into two distinct categories:

Of course, Matplotlib supports both the categories, particularly with the following output formats:

PS or EPS formats are particularly useful for plots inclusion in LaTeX documents, the main scientific articles format since decades.

In the previous section, we saw the file output formats — they are also called hardcopy backends as they create something (a file on disk).

A backend that displays the image on screen is called a user interface backend.

The backend is that part of Matplotlib that works behind the scenes and allows the software to target several different output formats and GUI libraries (for screen visualization).

In order to be even more flexible, Matplotlib introduces the following two layers structured (only for GUI output):

The standard renderer is the Anti-Grain Geometry ( AGG) library, a high performance rendering engine which is able to create images of publication level quality, with anti-aliasing, and subpixel accuracy. AGG is responsible for the beautiful appearance of Matplotlib graphs.

The canvas is provided with the GUI libraries, and any of them can use the AGG rendering, along with the support for other rendering engines (for example, GTK+).

Let's have a look at the user interface toolkits and their available renderers:

Here is the list of renderers for file output:

The renderers mentioned in the previous table can be used directly in Matplotlib, when we want only to save the resulting graph into a file (without any visualization of it), in any of the formats supported.

We have to pay attention when choosing which backend to use. For example, if we don't have a graphical environment available, then we have to use the AGG backend (or any other file). If we have installed only the GTK+ Python bindings, then we can't use the WX backend.

The following tools are needed if we're going to install Matplotlib from the source:

The advantage of using a Linux distribution is that several programs and libraries are already prepared by the distribution developers and made available (in a package format) to users. All we have to do is use the right tool and install the package.

In the following table, we will present some of the common Linux distributions package names for Matplotlib and the tools we can use to install the package:

Before we can install Matplotlib, we have to satisfy its main dependencies. So, we have to download:

Once we've got the above packages correctly installed, we can go to the main project page of Matplotlib on SourceForge at http://sourceforge.net/projects/matplotlib/. In the Files section, we can find the relative versions of the binary packages for the Python that we have just installed (2.4, 2.5, or 2.6).

The procedure to install Matplotlib correctly on Mac OS X is similar to that of Windows.

First of all, we need to download:

At this point, once they are correctly installed, we can download the binary installer from the download area of Matplotlib SourceForce page at http://sourceforge.net/projects/matplotlib/ or we can retrieve the version available at http://pythonmac.org/.

There are some packaged distributions of Python that contain Matplotlib in them, along with many other tools, such as IPython, NumPy, SciPy, and so on. These distributions will set up all the necessary things we need so that we can use Matplotlib on our machine. Some of the distributions are as follows:

These are mainly scientific distributions that install a lot of tools we don't directly need or use, but they have the advantage of making it easy to get Python, NumPy, and Matplotlib installed and working on our system.

There are two ways of obtaining the Matplotlib source code. They are:

If we decided to go with SVN, we can follow the instructions available in the Develop section of http://sourceforge.net/projects/matplotlib/.

If we are going to use the source code tarball, we will have to unpack it, go into the created source directory, and execute the following commands:

$ python setup.py build
$ sudo python setup.py install

These commands will build and then install Matplotlib. We will need administrative privileges to install it into the system directories (hence the sudo command in this Linux example).

Many aspects of the installation can be tuned using setup.cfg, a file shipped with the source code and used at build and install time. We can use it to customize the build process, such as changing the default backend, or choosing whether to install the optional libraries or not.

If we want to install Matplotlib from source on Windows, the Files section of Matplotlib SourceForge page contains handy egg files which we can download (choosing the Python version of interest) and then install using setuptools command. The following command will install Matplotlib on your machine:

$ easy_install matplotlib-<version>-py<py version>-win32.egg

Egg files are also available for Mac OS X, and we can use them in the same way as described above.

To ensure we have correctly installed Matplotlib and its dependencies, a very simple test can be carried out in the following manner:

$ python
Python 2.5.4 (r254:67916, Feb 18 2009, 03:00:47)
[GCC 4.3.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy
>>> print numpy.__version__
1.2.1
>>> import matplotlib
>>> print matplotlib.__version__
0.98.5.3

If there's no error while executing this, then we are done.