Updating the documentation

@author: Vincent Marois

It is of high priority that the documentation of MI Prometheus is kept up-to-date as the code base evolves. Good code without good documentation is not useful!

Guidelines & examples

Here is a quick how-to guide on how to keep this documentation up-to-date.

  • The documentation source files are contained in mi-prometheus/docs/source. This directory contains:

    • conf.py: Configuration file for the Sphinx documentation builder.
    • index.rst: master table of content document for the entire documentation.
    • models.rst: master table of content document for the models package.
    • problems.rst: master table of content document for the problems package.
    • utils.rst: master table of content document for the utils package.
    • workers.rst: master table of content document for the workers package.
    • notes/: contains global information about the documentation pages (e.g. this page).

Other folders (e.g. tutorials/) should be coming as the documentation grows.

You should not have to edit the conf.py file when doing changes to the code base. We mainly have to maintain the .rst files and the import lines in the __init__.py files.

The .rst files are written using the reStructuredText plaintext markup syntax. This is where you can define the hierarchy of the table of content. Here is an example of hierarchy:

Models  # This is a title

.. automodule:: miprometheus.models

Model  # this is a subtitle

.. autoclass:: Model

:hidden:`CNN + LSTM` # This is a subsubtitle
.. automodule:: miprometheus.models.vqa_baselines.cnn_lstm

SequentialModel # this is a subtitle
.. autoclass:: SequentialModel

:hidden:`DWM` # this is a subsubtitle
.. automodule:: miprometheus.models.dwm

Do not hesitate to frequently refer to the reStructuredText guide for more information on the formatting.

When adding a new module (.py file), class or function in the code base, please do the following:

  • First, update the closest __init__.py file in the mi-prometheus code hierarchy to import the new structures you have written. For instance, let’s assume you have written a file master_algorithm.py that contains a class named MeaningOfLife, and that this file is located in dir1/dir2/. In the __init__.py file of dir2/, add the following lines:

    >>> from .master_algorithm import MeaningOfLife
    >>> __all__ = [..., 'MeaningOfLife']

    The first line imports the new class from the file you wrote. The second line adds it to the list of public objects of that module, so that in the __init__.py file of dir1/, the line:

    >>> from .dir2 import *

    will properly import MeaningOfLife.

    So, if you are adding one .py file (called a module) to a directory that already contains a __init__.py file (this dir is then called a package), you only have to edit this __init__.

    If you are adding several directories and subdirectories, you have to update the several __init__.py files by traversing the hierarchy from innermost to outermost (simplest way to ensure you are not forgetting anything).

    NOTE: This may change in the future if our guidelines on the import lines change.

  • Second, we have to update the corresponding .rst file to include this new module/class in the table of content.

    The main markers (called directives by the reStructuredText syntax) are

    .. automodule:: dir1.dir2.master_algorithm


    .. autoclass:: dir1.dir2.master_algorithm.MeaningOfLife

    If you are adding only functions / methods within an existing class or module and if the reference to that class or module already exists in the table of content, you should not have to edit anything. Reconstructing the .html pages from the .rst files should automatically pull the corresponding docstrings.

    If you are adding a new class or module, then you have to add the reference in the table of content. The current (root) module in the .rst file should be indicated at the top of the file with the directive:

    .. currentmodule:: dir1

    So you just have to add the above sections at the location you want in the table of content hierarchy.

  • Finally, we have to rebuild the .html pages from the .rst files. This is done by readthedocs when we do a commit to our repository.

NOTE: We are not using the setup.py to build the documentation, but rather using mocking to ignore the dependencies. The reason is as follows:

  • The installation of the framework (through python setup.py install) can be resource intensive and the docker backend of readthedocs is constrained in terms of memory.
  • The documentation build should be pretty fast. Hence, avoiding dealing with dependencies is better.

Please refer to the readthedocs.yml file to see the configuration for the documentation build.

Some quotes about Code Documentation

"Always code as if the guy who ends up maintaining your code will be a violent psychopath who knows where you live." - John F. Woods
"Ink is better than the best memory." - Chinese proverb
"The documentation needs documentation." - a Bellevue Linux Users Group member, 2005