Ford

In this version:

• Documentation is much better
• Support for more layers and model configurations (including conditional GANs)
• More examples and bug fixes (recurrent attention wrappers work correctly)

Still in progress:

• User testing (to see how intuitive it is to use)
• Building better examples
• Improving documentation
• Adding more features

Arthur

Barebones implementation which supports normal GANs and ACGANs.

• Added xor.py and mnist.py examples, which illustrate supervised and unsupervised GAN training
• Basic model is in place
• No unit test (yet), but working towards it
• Made sure the XOR example works correctly (need access to a GPU to try the MNIST example)

Next steps:

• Work out bugs / unknown behavior: Model saving, fit_generator and similar.
• Try out different models to see what can be done
• Write more documentation: Document the layers better.