• Author : Jorge Herrera (jorgeh@ccrma.stanford.edu)
• Date : May, 2014

A python implementation of Gradient Frequency Neural Networks or GrFNNs (pronounced Griffins), introduced by Large et. al.. Dr. Large released a Matlab toolbox with his implementation (see here). pyGrFNN was developed in parallel but in close collaboration with Dr. Large, so while there are similarities, there are are some differences as well.

NOTE This package is in development, so things may or may not change in the future.

You can find the documentation here. It includes some examples.

Clone the repo to a local directory. Inside the director run

python setup.py develop

This will install the package in development mode, so every time you pull changes to the repo, they should be automatically reflected in the installed version.

Alternatively, you could run

python setup.py install

which performs a normal installation.

In either case, if all goes well, all dependencies should be installed.

This package has been developed and tested exclusively on Python 2.7. Other that that, you will need:

• numpy
• scipy
• PyDispatcher

Optional dependencies:

• matplotlib (to enable custom visualizations)
• nose (for testing)
• sphinx (to build docs)
• sphinxcontrib-napoleon (to build docs)
• [sphinx_rtd_theme](http://read-the-docs.readthedocs.org/en/latest/theme. html) (to build docs)

(all dependencies are available via pip)

You can find the documentation here.

Documentation is generated with Sphinx directly from the docstrings in the code. The theme is the ReadTheDocs theme, that can be found here (it includes instructions on how to install and use it)

cd <pygrfnn_dir>/docs
make html
open _build/html/index.html

The last line is an OS X way of opening a file on a browser from the command line. On other systems, simply open file://<absolute_path_to_pygrfnn>/docs/_build/html/index.html.

If something goes wrong, you can optionally run make clean to build from scratch.

This repo was intended to use virtualenv, but for some reason, virtualenv and matplotlib decided to not like each other, so in the mean time I'm using the system-site-packages (at least for NumPy, SciPy and Matplotlib). That is on my local environment, but you are encouraged to use virtualenv if possible.

Testing uses nose (pip install nose). To run all the tests, simply type:

nosetests -v

in the root folder (where pygfnn and test folders are). The (optional) -v flag means verbose. To run only unit or functional test, use the -w option:

nosetests -w ./ ./test/unit