Advanced usage

This page contains more advanced topics in case you want to understand how to use Sphinx-Gallery more deeply.

Write a custom image scraper

Warning

The API for custom scrapers is currently experimental.

By default, Sphinx-gallery supports image scrapers for Matplotlib (matplotlib_scraper()) and Mayavi (mayavi_scraper()). You can also write a custom scraper for other python packages. This section describes how to do so.

Image scrapers are functions (or callable class instances) that do two things:

  1. Collect a list of images created in the latest execution of code.
  2. Write these images to disk in PNG, JPEG, or SVG format (with .png, .jpg, or .svg extensions, respectively)
  3. Return rST that embeds these figures in the built documentation.

The function should take the following inputs (in this order):

  1. block - a Sphinx-Gallery .py file is separated into consecutive lines of ‘code’ and rST ‘text’, called ‘blocks’. For each block, a tuple containing the (label, content, line_number) (e.g. ('code', 'print("Hello world")', 5)) of the block is created.
    • ‘label’ is a string that can either be 'text' or 'code'. In this context, it should only be 'code' as this function is only called for code blocks.
    • ‘content’ is a string containing the actual content of the code block.
    • ‘line_number’ is an integer, indicating the line number that the block starts at.
  2. block_vars - dictionary of configuration and runtime variables. Of interest for image scrapers is the element 'image_path_iterator' which is an iterable object which returns an absolute path to an image file name adhering to Sphinx-Gallery naming convention. The path directs to the gallery_dirs/images directory (Configure and use sphinx-gallery) and the image file name is 'sphx_glr_' followed by the name of the source .py file then a number, which starts at 1 and increases by 1 at each iteration. The default file format is .'png'. For example: 'home/user/Documents/module/auto_examples/images/sphx_glr_plot_mymodule_001.png'
  3. gallery_conf - dictionary containing the configuration of Sphinx-Gallery, set under sphinx_gallery_conf in doc/conf.py (Configuration).

It should return a string containing the rST for embedding this figure in the documentation. See matplotlib_scraper() for an example of a scraper function (click on ‘source’ below the function name to see the source code). The matplotlib_scraper() uses the helper function sphinx_gallery.scrapers.figure_rst() to help generate rST (see below).

This function will be called once for each code block of your examples. Sphinx-gallery will take care of scaling images for the gallery index page thumbnails. PNG images are scaled using Pillow, and SVG images are copied.

Warning

SVG images do not work with latex build modes, thus will not work while building a PDF vesion of your documentation.

Example 1: a Matplotlib and Mayavi-style scraper

For example, we will show sample code for a scraper for a hypothetical package. It uses an approach similar to what sphinx_gallery.scrapers.matplotlib_scraper() and sphinx_gallery.scrapers.mayavi_scraper() do under the hood, which use the helper function sphinx_gallery.scrapers.figure_rst() to create the standardized rST. If your package will be used to write an image file to disk (e.g., PNG or JPEG), we recommend you use a similar approach.

def my_module_scraper(block, block_vars, gallery_conf)
    import mymodule
    # We use a list to collect references to image names
    image_names = list()
    # The `image_path_iterator` is created by Sphinx-gallery, it will yield
    # a path to a file name that adheres to Sphinx-gallery naming convention.
    image_path_iterator = block_vars['image_path_iterator']

    # Define a list of our already-created figure objects.
    list_of_my_figures = mymodule.get_figures()

    # Iterate through figure objects, save to disk, and keep track of paths.
    for fig, image_path in zip(list_of_my_figures, image_path_iterator):
        fig.save_png(image_path)
        image_names.append(image_path)

    # Close all references to figures so they aren't used later.
    mymodule.close('all')

    # Use the `figure_rst` helper function to generate the rST for this
    # code block's figures. Alternatively you can define your own rST.
    return figure_rst(image_names, gallery_conf['src_dir'])

This code would be defined either in your conf.py file, or as a module that you import into your conf.py file. The configuration needed to use this scraper would look like:

sphinx_gallery_conf = {
    ...
    'image_scrapers': ('matplotlib', my_module_scraper),
}

Example 2: detecting image files on disk

Here’s another example that assumes that images have already been written to disk. In this case we won’t generate any image files, we’ll only generate the rST needed to embed them in the documentation.

We’ll use a callable class in this case, and assume it is defined within your package in a module called scraper. Here is the scraper code:

from glob import glob
import shutil
import os
from sphinx_gallery.gen_rst import figure_rst

class PNGScraper(object):
    def __init__(self):
        self.seen = set()

    def __repr__(self):
        return 'PNGScraper'

    def __call__(self, block, block_vars, gallery_conf):
        # Find all PNG files in the directory of this example.
        path_current_example = os.path.dirname(block_vars['src_file'])
        pngs = sorted(glob(os.path.join(os.getcwd(), '*.png'))

        # Iterate through PNGs, copy them to the sphinx-gallery output directory
        image_names = list()
        image_path_iterator = block_vars['image_path_iterator']
        for png in pngs:
            if png not in seen:
                seen |= set(png)
                this_image_path = image_path_iterator.next()
                image_names.append(this_image_path)
                shutil.move(png, this_image_path)
        # Use the `figure_rst` helper function to generate rST for image files
        return figure_rst(image_names, gallery_conf['src_dir'])

Then, in our conf.py file, we include the following code:

from mymodule import PNGScraper

sphinx_gallery_conf = {
    ...
    'image_scrapers': ('matplotlib', PNGScraper()),
}

Example 3: matplotlib with SVG format

The sphinx_gallery.scrapers.matplotlib_scraper() supports **kwargs to pass to matplotlib.figure.Figure.savefig(), one of which is the format argument. Currently sphinx-gallery supports PNG (default) and SVG output formats. To use SVG, you can do:

from sphinx_gallery.scrapers import matplotlib_scraper

class matplotlib_svg_scraper(object):

    def __repr__(self):
        return self.__class__.__name__

    def __call__(self, *args, **kwargs):
        return matplotlib_scraper(*args, format='svg', **kwargs)

sphinx_gallery_conf = {
    ...
    'image_scrapers': (matplotlib_svg_scraper(),),
    ...
}

You can also use different formats on a per-image basis, but this requires writing a customized scraper class or function.

Define resetting behavior (e.g., for custom libraries)

Sphinx-gallery natively supports resetting matplotlib and seaborn. However, if you’d like to support resetting for other libraries (or would like to modify the resetting behavior for a natively-supported library), you can add a custom function to the resetting tuple defined in conf.py.

The function takes two variables: a dictionary called gallery_conf (which is your Sphinx-gallery configuration) and a string called fname (which is the file name of the currently-executed Python script). These generally don’t need to be used in order to perform whatever resetting behavior you want, but must be included in the function definition for compatibility reasons.

For example, to reset matplotlib to always use the ggplot style, you could do:

def reset_mpl(gallery_conf, fname):
    from matplotlib import style
    style.use('ggplot')

Any custom functions can be defined (or imported) in conf.py and given to the reset_modules configuration key. For the function defined above:

sphinx_gallery_conf = {
    ...
    'reset_modules': (reset_mpl, 'seaborn'),
}

Note

Using resetters such as reset_mpl that deviate from the standard behavior that users will experience when manually running examples themselves is discouraged due to the inconsistency that results between the rendered examples and local outputs.